probabilistic leak detection in pipelines using the mass imbalance approach

Applicable to a Wide Range of Pipe Diameters.Capacitive Sensors unaffected by liquid color.•Mount to bypass pipes.•Fit a wide range of pipe diameters: 8 to 11 mm or 12 to 26mm •Built-in Amplifiers to save space.Be sure to read Safety Precautionsonpage 3.Ordering InformationSensing methodApplicable pipe diametersAppearanceOutput configuration/Operation modeModelCapacitive 8 to 11 mm NPN open-collector output NOE2K-L13MC1 2M12 to 26 mm E2K-L26MC1 2ME2K-LRatings and Specifications*Stable detection will not be possible in the following cases. Confirm detection capability with the Sensor installed before actual application.1. If the specific inductive capacity or the specific electric conductivity of the liquid is too low, the liquid position may not be detected since this sensor is a capacitive sensor.2. If the quantity of liquid is too low or the change in quantity is too low in comparison to the change in liquid level because the pipe is too thin or the walls of the pipe are too thick3. If there is a viscous film on the inner pipe wall, large quantities of foam or air bubbles, or excessive buildup of dirt on the inner pipe wallEngineering Data (Typical)ItemModelE2K-L13MC1E2K-L26MC1Applicable pipesMaterialsNon-metal SizeDiame-ter 8 to 11 mm 12 to 26 mm Thick-ness 1 mm max. 1.5 mm max.Detectable object Liquid *Repeat accuracy±0.2 mm max.Differential travel(Reference value, varies with pipe size and liquid.)0.6 to 5 mm 0.3 to 3 mm Power supply voltage (operating voltage range)12 to 24 VDC (10.8 to 30 VDC), ripple (p-p): 10% max.Current consumption 12 mA max.Control outputLoad current100 mA max.Residual voltage 1 V max. (Load current: 100 mA, Cable length: 2 m)Sensing liquid position Indented mark position (For details, refer to Technical Guide (Operational version).)IndicatorsDetection indicator (orange)Ambient temperature range Operating: 0 to 55°C (with no icing or condensation), Storage: −10 to 65°C (with no icing or condensation)Ambient humidity range Operating/Storage: 25% to 85% (with no condensation)Temperature influence ±4 mm of detection level at 23°C in the temperature range of 0 to 55°C (with pure water or 20% saline solution)(±6 mm for E2K-L13MC1 with pure water and a pipe diameter of 8 mm)Voltage influence ±0.5 mm of detection level at the rated voltage in rated voltage ±10% range Insulation resistance 50 M Ω min. (at 500 VDC) between current-carrying parts and case Dielectric strength 500 VAC, 50/60 Hz for 1 min between current-carrying parts and caseVibration resistance Destruction: 10 to 55 Hz, 1.5-mm double amplitude for 2 hours each in X, Y, and Z directions Shock resistance Destruction: 500 m/s 2 3 times each in X, Y, and Z directions Degree of protection IP66 (IEC)Connection method Pre-wired Models (Standard cable length: 2 m)Weight (packed state)Approx. 70 g Materials Case, Cover Heat-resistant ABS Cable clampNBRAccessoriesTwo bands, Four slip-proof tubes, Instruction manualInfluence of Temperature and Sensing Liquid Level E2K-L13MC1E2K-L26MC1−−−C h a n g e i n s e n s i n g l i q u i d l e v e l (m m )−−−C h a n g e i n s e n s i n g l i q u i d l e v e l (m m )ratings.● DesignInfluence of Surrounding ObjectsWhen mounting the Sensor, maintain at least the distances in the following diagrams from surrounding metal objects or otherconductors to prevent the Sensor from being affected by objects other than the sensing object.Influence of Surrounding Objects(Unit: mm)Mutual InterferenceWhen installing Sensors in series, in parallel, or face-to-face, ensure that the minimum distances given in the following table are maintained.Sensor may affect the detection level of the top Sensor.● Mounting MountingMount the Sensor securely to the pipe using the enclosed two bands and four slip-proof tubes (two tubes used for each band) as shown in the following diagram.When mounting the Sensor, be sure the entire Sensor is tight against the pipe along the sensing surface.ModelDistanceA B C E2K-L13MC125545E2K-L26MC140On One Side Liquid levelSet position mark)PipeSlip-proof tubesBandsSide ViewFront ViewTop ViewE2K-L● WiringPower Supply•If the load and Sensor are connected to different power supplies, always turn ON the Sensor power first.•Switching noise can cause operating mistakes if a commercial switching regulator is used. When using a switching regulator, always ground the frame ground terminal and the ground terminal. ● Operating EnvironmentAmbient Atmosphere•Although the Sensor is water resistance, it is a capacitive sensor and should not be used where it will come into direct contact with liquids, such as water or cutting oil.•The life of the Sensor will be shorten by rapid changes in temperature even within the ambient operating temperature range. Do not use the Sensor in locations subject to rapid temperature changes.● MiscellaneousDrift will occur when the power supply is turned ON. If the specific inductive capacity of the sensing liquid is low, the detection level may increase by 2 to 3 mm during the 20 minutes required from the time the power supply is turned ON until operation stabilizes.Dimensions (Unit: mm)Tolerance class IT16 applies to dimensions in this data sheet unless otherwise specified.E2K-L13MC1E2K-L26MC1Read and Understand This CatalogPlease read and understand this catalog before purchasing the products. Please consult your OMRON representative if you have any questions orcomments.WARRANTYOMRON's exclusive warranty is that the products are free from defects in materials and workmanship for a period of one year (or other period if specified) from date of sale by OMRON.OMRON MAKES NO WARRANTY OR REPRESENTATION, EXPRESS OR IMPLIED, REGARDING NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR PARTICULAR PURPOSE OF THE PRODUCTS. ANY BUYER OR USER ACKNOWLEDGES THAT THE BUYER OR USER ALONE HAS DETERMINED THAT THE PRODUCTS WILL SUITABLY MEET THE REQUIREMENTS OF THEIR INTENDED USE. OMRON DISCLAIMS ALL OTHER WARRANTIES, EXPRESS OR IMPLIED.LIMITATIONS OF LIABILITYOMRON SHALL NOT BE RESPONSIBLE FOR SPECIAL, INDIRECT, OR CONSEQUENTIAL DAMAGES, LOSS OF PROFITS OR COMMERCIAL LOSS IN ANY WAY CONNECTED WITH THE PRODUCTS, WHETHER SUCH CLAIM IS BASED ON CONTRACT, WARRANTY, NEGLIGENCE, OR STRICT LIABILITY.In no event shall the responsibility of OMRON for any act exceed the individual price of the product on which liability is asserted.IN NO EVENT SHALL OMRON BE RESPONSIBLE FOR WARRANTY, REPAIR, OR OTHER CLAIMS REGARDING THE PRODUCTS UNLESSOMRON'S ANALYSIS CONFIRMS THAT THE PRODUCTS WERE PROPERLY HANDLED, STORED, INSTALLED, AND MAINTAINED AND NOTSUBJECT TO CONTAMINATION, ABUSE, MISUSE, OR INAPPROPRIATE MODIFICATION OR REPAIR.SUITABILITY FOR USEOMRON shall not be responsible for conformity with any standards, codes, or regulations that apply to the combination of products in the customer's application or use of the products.At the customer's request, OMRON will provide applicable third party certification documents identifying ratings and limitations of use that apply to the products. This information by itself is not sufficient for a complete determination of the suitability of the products in combination with the end product, machine, system, or other application or use.The following are some examples of applications for which particular attention must be given. This is not intended to be an exhaustive list of all possible uses of the products, nor is it intended to imply that the uses listed may be suitable for the products:•Outdoor use, uses involving potential chemical contamination or electrical interference, or conditions or uses not described in this catalog.•Nuclear energy control systems, combustion systems, railroad systems, aviation systems, medical equipment, amusement machines, vehicles, safety equipment, and installations subject to separate industry or government regulations.•Systems, machines, and equipment that could present a risk to life or property.Please know and observe all prohibitions of use applicable to the products.NEVER USE THE PRODUCTS FOR AN APPLICATION INVOLVING SERIOUS RISK TO LIFE OR PROPERTY WITHOUT ENSURING THAT THE SYSTEM AS A WHOLE HAS BEEN DESIGNED TO ADDRESS THE RISKS, AND THAT THE OMRON PRODUCTS ARE PROPERLY RATED AND INSTALLED FOR THE INTENDED USE WITHIN THE OVERALL EQUIPMENT OR SYSTEM.PROGRAMMABLE PRODUCTSOMRON shall not be responsible for the user's programming of a programmable product, or any consequence thereof.CHANGE IN SPECIFICATIONSProduct specifications and accessories may be changed at any time based on improvements and other reasons.It is our practice to change model numbers when published ratings or features are changed, or when significant construction changes are made.However, some specifications of the products may be changed without any notice. When in doubt, special model numbers may be assigned to fix or establish key specifications for your application on your request. Please consult with your OMRON representative at any time to confirm actualspecifications of purchased products.DIMENSIONS AND WEIGHTSDimensions and weights are nominal and are not to be used for manufacturing purposes, even when tolerances are shown.PERFORMANCE DATAPerformance data given in this catalog is provided as a guide for the user in determining suitability and does not constitute a warranty. It may represent the result of OMRON’s test conditions, and the users must correlate it to actual application requirements. Actual performance is subject to the OMRON Warranty and Limitations of Liability.ERRORS AND OMISSIONSThe information in this document has been carefully checked and is believed to be accurate; however, no responsibility is assumed for clerical,typographical, or proofreading errors, or omissions.2008.11In the interest of product improvement, specifications are subject to change without notice. OMRON CorporationIndustrial Automation Company/(c)Copyright OMRON Corporation 2008 All Right Reserved.分销商库存信息:OMRONE2K-L26MC1E2K-L13MC1。

Eaton Tedeco® brand Chip Detectors are widely used to provide early failure detection in the lube systems of helicopter engines, transmissions and auxiliary power units, and, to a lesser extent,in commercial or military gas turbine engine lubesystems.They consist of a magnet surrounded by two closely spaced conductive pole shoes that are insulated from each other. Ferrous debris particles, captured by the magnet, eventually bridge the pole shoe gap, completing a conductive circuit similar to an electric switch. They are typically series connected through a power supply and an indicating lamp in the cockpit of an aircraft. When combined with a self-closing valve, the chip detector can be removed for inspection or cleaning while keeping oil loss at a minimum. Removal of the plug is facilitated by the quick-disconnect Helilok® mechanism. A simple pushing motion, which the helical cam converts into rotation, unlocks the unit while a strong spring pushes it out, at the same time closing a valve. This mechanism assures positive locking when the Options:Drain and samplingyattachments Features:Terminal configurations yinclude threaded bayonet hermetic receptacles or stud terminalsHeliloky®quick-disconnect and positive locking mechanism Lockwiring of chipydetectorplug not required Retains ferrous debris yoff-line analysisSelf-closing valve allows yremoval of chip detector with minimum oil loss Scalloped hand gripypermits a secure hold on plug, even with arctic glovesplug is returned since the helical locking cams have only one stable position, that is, when they are locked.Lockwiring of the plug is not required.AS 9100 Chip Detectors with Self-Closing Valves©2009 Eaton CorporationAll Rights ReservedPrinted In USACopying or Editing is Forbidden Form No. DS200-3BJuly 2009EatonAerospace Group Electrical Sensing & Controls 24 East Glenolden Avenue Glenolden, PA 19036-2198 USAtel: (610) 522-4000fax: (610) 522-4900/aerospaceChip Detector Contact Gap DesignsDepending on wear, particle size, and quantity, chip detectors are available with either axial or radial gap configuration.Axial Gap Chip DetectorsThe axial gap chip detector has two pole pieces which have a gap between them in an axial direction relative to the magnetic chip detector. Characterized by improved contact exposure, high particle retention and increased smallparticle sensitivity, this type is ideally suited to engine lube systems, high speed pumps and other equipment with high surface speeds that generate small particles.Radial Gap Chip Detectors The radial gap chip detector has two pole pieces which have a gap between them in a radial direction relative to the magnetic chip detector. The pole pieces are located at the end of the magnet nearest the oil stream and have an annulus extending radially outward from the centerline of the detector. This type is recommended for helicopter gearboxes and transmissions and other relatively low-speed power trains where impending failure produces wear particles of larger size and quantity.Axial Gap Chip DetectorRadial Gap Chip DetectorContact Gap Contact Gap。

匹配场处理检测因子下载温馨提示：该文档是我店铺精心编制而成，希望大家下载以后，能够帮助大家解决实际的问题。

文档下载后可定制随意修改，请根据实际需要进行相应的调整和使用，谢谢!并且，本店铺为大家提供各种各样类型的实用资料，如教育随笔、日记赏析、句子摘抄、古诗大全、经典美文、话题作文、工作总结、词语解析、文案摘录、其他资料等等，如想了解不同资料格式和写法，敬请关注!Download tips: This document is carefully compiled by the editor. I hope that after you download them, they can help yousolve practical problems. The document can be customized and modified after downloading, please adjust and use it according to actual needs, thank you!In addition, our shop provides you with various types of practical materials, such as educational essays, diary appreciation, sentence excerpts, ancient poems, classic articles, topic composition, work summary, word parsing, copy excerpts,other materials and so on, want to know different data formats and writing methods, please pay attention!匹配场处理检测因子在现代科学研究中扮演着重要角色，通过匹配场处理检测因子的测定，可以更准确地分析样品中的成分和含量。

Toxoplasma IgG ELISA Catalog Number SE120126Storage Temperature 2–8 °CTECHNICAL BULLETINProduct DescriptionToxoplasma gondii causes toxoplasmosis, a common disease that affects 30–50 of every 100 people in North America by the time they are adults. The mean source of infection is direct contact with cat feces or from eating undercooked meats. Toxoplasmosis generally presents with mild symptoms in immunocompetent individuals. In the immunocompromised patient,however, the infection can have serious consequences. Acute toxoplasmosis in pregnant women can result in miscarriage, poor growth, early delivery, or stillbirth. Treatment of an infected pregnant woman may prevent or lessen the disease in her unborn child. Treatment of an infected infant will also lessen the severity of the disease as the child grows. IgG and IgM antibodies to Toxoplasma can be detected with 2–3 weeks after exposure. IgG remains positive, but the antibody level drops overtime. ELISA can detect Toxoplasma IgM antibody one year after infection in over 50% of patients. Therefore, IgM positive results should be evaluated further with one or two follow up samples if primary infection is suspected.The Toxoplasma IgG ELISA Kit is intended for the detection of IgG antibody to Toxoplasma in human serum or plasma. Diluted serum is added to wellscoated with purified Toxoplasma antigen. Toxoplasma IgG specific antibody, if present, binds to the antigen. All unbound materials are washed away and the enzyme conjugate is added to bind to the antibody-antigen complex, if present. Excess enzyme conjugate is washed off and substrate is added. The plate isincubated to allow the oxidation of the substrate by the enzyme. The intensity of the color generated isproportional to the amount of IgG specific antibody in the sample.ComponentsReagents and Equipment Required but Not Provided.1.Distilled or deionized water2.Precision pipettes3.Disposable pipette tips4.ELISA reader capable of reading absorbance at450 nm5.Absorbent paper or paper towel6.Graph paperPrecautions and DisclaimerThis product is for R&D use only, not for drug,household, or other uses. Please consult the Safety Data Sheet for information regarding hazards and safe handling practices.Preparation Instructions Sample Preparation1. Collect blood specimens and separate the serum.2.Specimens may be refrigerated at 2–8°C for up toseven days or frozen for up to six months. Avoid repetitive freezing and thawing.20x Wash Buffer ConcentratePrepare 1x Wash buffer by adding the contents of the bottle (25 mL, 20x) to 475 mL of distilled or deionized water. Store at room temperature (18–26 °C).2Storage/StabilityStore the kit at 2–8 °C.ProcedureNotes: The components in this kit are intended for use as an integral unit. The components of different lots should not be mixed.Optimal results will be obtained by strict adherence to the test protocol. Precise pipetting as well as following the exact time and temperature requirements is essential.The test run may be considered valid provided the following criteria are met:1.If the O.D. of the Calibrator is >0.250.2.The Ab index for Negative control should be <0.9.3.The Ab index for Positive control should be >1.2. Bring all specimens and kit reagents to room temperature (18–26 °C) and gently mix. 1.Place the desired number of coated strips into theholder.2.Negative control, positive control, and calibrator areready to use.Prepare 21-fold dilution of testsamples, by adding 10 µL of the sample to 200 µLof Sample Diluent. Mix well.3.Dispense 100 µL of diluted sera, calibrator, andcontrols into the appropriate wells. For the reagent blank, dispense 100 µL of Sample Diluent in 1Awell position. Tap the holder to remove air bubbles from the liquid and mix well. Incubate for20 minutes at room temperature.4.Remove liquid from all wells. Wash wells threetimes with 300 µL of 1x wash buffer. Blot onabsorbent paper or paper towel.5.Dispense 100 µL of enzyme conjugate to each welland incubate for 20 minutes at room temperature. 6.Remove enzyme conjugate from all wells. Washwells three times with 300 µL of1x wash buffer.Blot on absorbent paper or paper towel7.Dispense 100 µL of TMB substrate and incubate for10 minutes at room temperature.8.Add 100 µL of stop solution.9.Read O.D. at 450 nm using ELISA reader within15 minutes. A dual wavelength is recommendedwith reference filter of 600–650 nm.3ResultsCalculations1.Check Calibrator Factor (CF) value on thecalibrator bottle. This value might vary from lot tolot. Make sure the value is checked on every kit. 2.Calculate cut-off value: Calibrator OD x CalibratorFactor (CF).3.Calculate the Ab (Antibody) Index of eachdetermination by dividing the mean values of each sample by cut-off value.Example of typical results:Calibrator mean OD = 0.8Calibrator Factor (CF) = 0.5Cut-off Value = 0.8 x 0.5 = 0.400Positive control O.D. = 1.2Ab Index = 1.2/0.4 = 3Patient sample O.D. = 1.6Ab Index = 1.6/0.4 = 4.0Note:Lipemic or hemolyzed samples may cause erroneous results.InterpretationThe following is intended as a guide to interpretation of Toxoplasma IgG antibody index (Ab Index) test results; each laboratory is encouraged to establish its own criteria for test interpretation based on sample populations encountered.<0.9 –No detectable IgG antibody to Toxoplasma byELISA0.9–1.1 –Borderline positive. Follow-up testing isrecommend if clinically indicated.>1.1 –Detectable IgG antibody to Toxoplasma by ELISA References1.Wilson, M. et al., Evaluation of six commercial kitsfor detection of human immunoglobulin Mantibodies to Toxoplasma gondii. The FDAToxoplasmosis Ad Hoc Working Group.2.Obwaller, A. et al., An enzyme-linkedimmunosorbent assay with whole trophozoites ofToxoplasma gondii from serum-free tissue culturefor detection of specific antibodies. Parasitol. Res., 1995;81(5):361-4.3.Loyola, A.M. et al., Anti-Toxoplasma gondiiimmunoglobulins A and G in human saliva andserum. J. Oral Pathol. Med., 1997; 26(4):187-91. 4.Doehring, E. et al., Toxoplasma gondii antibodies inpregnant women and their newborns in Dar esSalaam, Tanzania. Am. J. Trop. Med. Hyg., 1995;52(6):546-8.5.Cotty, F. et al., Prenatal diagnosis of congenitaltoxoplasmosis: the role of Toxoplasma IgAantibodies in amniotic fluid [letter]. J. Infect. Dis.,1995;171(5):1384-5.6.Altintas, N. et al., Toxoplasmosis in last four yearsin Agean region,Turkey. J. Egypt. Soc. Parasitol.,1997;27(2):439-43.RGC,CH,MAM 10/14-1©2014 Sigma-Aldrich Co. LLC. All rights reserved. SIGMA-ALDRICH is a trademark of Sigma-Aldrich Co. LLC, registered in the US and other countries. Sigma brand products are sold through Sigma-Aldrich, Inc. Purchaser must determine the suitability of the product(s) for their particular use. Additional terms and conditions may apply. Please see product information on the Sigma-Aldrich website at and/or on the reverse side of the invoice or packing slip.。

VLI-880, VLI-885The VESDA VLI by Xtralis is an industry first early warning aspirating smokedetection (ASD) system, designed to protect industrial applications and harshenvironments of up to 2,000 m2 (21,520 sq. ft.).Long Life, Intelligent, Fail-Safe TechnologyThe VLI detector combines a fail-safe Intelligent Filter (patent pending) withan advanced clean-air barrier for optics protection allowing the use of absolutedetection and a long detection chamber life without the need for re-calibration.The Intelligent Filter:• Reduces the level of pollution in the air sample before it enters thedetection chamber, which dramatically extends the operational life of thedetector in harsh and polluted environments.• Is fully monitored, providing consistent sensitivity over the entire operational life of the detector. Installation, Commissioning and OperationThe VLI detector features a robust IP66-rated enclosure which provides complete protection against dust ingress and strong water jets from all directions. In the majority of industrial applications, specifically in very harsh environments, this eliminates the need to use expensive external IP enclosures, thus simplifying and reducing the cost of installation.The VLI detector is equipped with a powerful aspirator that provides a total pipe length of 360 m (1,181 ft). It is fully supported by the Xtralis ASPIRE, VSC and VSM4 software applications which facilitate ease of pipe network design, system commissioning and maintenance together with compatibility with existing VESDA installations.The AutoLearn™ commissioning assistant reduces setup time and ensures optimum alarm and flow thresholds in a range of environments.The VLI detector is inherently less prone to nuisance alarms due to the intelligent filter, lint trap, sub-sampling probe and secondary filter. Coupled with its modular design, VLI offers a lower total cost of ownership over the life of the product.Features• Suitable for Class 1 Division 2 applications - Groups A, B, C & D• Up to 2,000 m2 (21,520 sq. ft.) coverage• Up to 4 inlet pipes• Total pipe length up to 360 m (1,181 ft)• Five (5) high intensity status LEDs for greater visibility• Robust absolute smoke detection• Intelligent Filter (patent pending)• Lint Trap to capture fibrous particulates• Sub-sampling Probe (inertial separator)• Secondary Filter• Clean air barrier for optics protection• Referencing• AutoLearn™ Smoke and Flow• Clean Air Zero™• Air-path monitoring• Five (5) relays (Fire, Fault and 3 configurable)• Relays configurable as latching or non-latching • Expandable GPI and relays• Ultrasonic flow sensing• Xtralis VSC, Xtralis VSM4 and ASPIRE software support• IP66 Enclosure • Easy mounting with steel support bracket• Modular field replaceable parts for ease of servicing • Local USB configuration port• Easy cable termination access• Imperial and metric pipe ports• Rubberized finish to external housingListings / Approvals• UL• ULC• FM• ActivFire• CE• UKCA• LPCB• NF• SIL 2 as per IEC 61508• EN 54-20-Class A (24 holes / Alert = 0.06% obs/m)-Class B (28 holes / Fire-1 = 0.15% obs/m)-Class C (60 holes / Fire-1 = 0.15% obs/m)Classification of any configuration is determined using ASPIRE. Regional approvals listings and regulatory compliance vary between productmodels. Refer to for the latest product approvals matrix.All technical data is correct at the time of publication and is subject to changes without notice. AllIntellectual Property including but not limited to trademarks, copyrights, patent are hereby acknowledged. You agree not to copy, communicate to the public, adapt, distribute, transfer, sell, modify or publish any contents of this document without the express prior written consent of Xtralis. Installation information: In order to ensure full functionality, refer to the installation instructions as supplied. © XtralisDoc. No. 17115_20 Part No. AD29414-001 August 2021Clean Air ZeroClean Air Zero is a user-initiated VLI feature which compliments consistent absolute detection over time and also safeguards against nuisance alarms.This is achieved by introducing clean air into thedetection chamber and taking a reference reading of the chamber background. This reading is then offset against the actual environmental background to maintain consistent absolute smoke detection.Ordering InformationOrdering CodeDescriptionVLI-880VESDA VLIVLI-885VESDA VLI with VESDAnet* VRT-Q00VESDA VLI Remote Display 7 Relays VRT-T00VESDA VLI Remote Display No Relays* Note: Please contact your nearest Xtralis office for availability.Spare PartsOrdering CodeDescriptionVSP-030VLI Intelligent Filter VSP-031VLI Secondary Foam Filter VSP-032VLI AspiratorVSP-033VLI Chamber Assembly VSP-034VLI VESDAnet CardVSP-035VLI Remote Display Module Spare VSP-036VLI Ultrasonic Flow Manifold SpareTECHNICAL SPECIFICATIONSHow It WorksAir is continually drawn through the pipe network and into the VLI detector by a high efficiency aspirator. The air sample passes four (4) sets of ultrasonic flow sensors before being passed through the Intelligent Filter. The Intelligent Filter incorporates an innovative flow splitting arrangement where a smaller unfiltered portion is passed through another set of ultrasonic flow sensors and a larger portion of the sample passes through a HEPA filtration medium. This arrangement dramatically reduces the amount of contaminants entering the aspirator and the detection chamber, thus extending detector life.Filter loading is constantly monitored which enables the detector to “intelligently” maintain the sensitivity, hence ensuringconsistent and reliable operation over time. This is achieved by comparing the readings from the four (4) sets of ultrasonic flow sensors at the detector air inlets to the readings from that in the unfiltered path and measuring the split of the airflow ratio as the filter load changes.The filtered and unfiltered portions are recombined as they exit the Intelligent Filter. A portion of the recombined air sample is then passed through the sub-sampling probe (inertial separator) and secondary filter. This ensures that larger dust particles are less likely to pass through the probe and filter arrangement, hence they are exhausted out of the detector. This configuration minimizes nuisance alarms caused by larger dust particles and extends detection chamber life. A third filter within the detection chamber assembly delivers a clean air barrier which protects the optical surfaces from contamination, further extending detector life and ensuring absolute calibration.The detection chamber uses a stable, highly efficient laser light source and unique sensor configuration to achieve optimum response to a wide range of smoke types. The presence of smoke in the detection chamber creates light scattering which is detected by the very sensitive sensor circuitry and then converted to an alarm signal.The status of the detector, all alarms, service and fault events, are monitored and logged with time and date stamps. Status reporting can be transmitted via relay outputs and across VESDAnet (VN version only).Supply Voltage 18-30 VDCPower Consumption 10 W quiescent, 10.5 W with alarm (max)Current Consumption 415 mA quiescent, 440 mA with alarm (max)Fuse Rating 1.6 ADimensions (WHD)426.5 mm x 316.5 mm x 180 mm (16.8 in x 12.5 in x 7.1 in)Weight6.035 kg (13.3 lbs)Operating ConditionsTested to -10°C to 55°C (14°F to 131°F)Recommended Detector Ambient: 0°C to 39°C Sampled Air: -20°C to 60°C (-4°F to 140°F) Humidity: 10% to 95% RH, non-condensing Sampling Network Maximum area of Coverage 2,000 m 2 (21,520 sq.ft) Minimum total airflow: 40 l/m Minimum airflow per pipe: 20 l/mMaximum Pipe LengthsTotal Pipe Length: 360 m (1,181 ft)Maximum Single Pipe Length: 120 m (394 ft)Computer Design Tool ASPIREPipe Internal Diameter 15 mm - 21 mm (9/16”–7/8”) External Diameter 25 mm (1”)Relays 5 Relays rated 2 A @ 30 VDCFire (NO), Fault (NC), Configurable (NO)IP Rating IP66Cable Access 4 x 25 mm (1”) cable entriesCable Termination Screw Terminal blocks 0.2–2.5 sq mm (30–12 AWG)Sensitivity Range 0.005% - 20.0% obs/m (0.0015% - 6.25% obs/ft)Threshold Setting RangeAlert: 0.05%-1.990% obs/m (0.016%-0.6218% obs/ft) Action: 0.1%-1.995% obs/m (0.031%-0.6234% obs/ft) Fire1: 0.15 %-2.0% obs/m (0.047% - 0.625% obs/ft) Fire2: 0.155 % - 20.0% obs/m (0.05% - 6.25% obs/ft)* * Limited to 4% obs/ft for ULSoftware FeaturesEvent log: Up to 18,000 events stored in FIFO formatSmoke level, user actions, alarms and faults with time and date stampAutoLearn: Min 15 minutes, Max 15 days. Recommended minimum 14 days.While AutoLearn is in progress, thresholds are NOT changed from pre-set values.Configurable General Input (5 - 30 VDC)External Reset, Mains OK, Standby, Isolate, Use Night-time Threshold, Reset + Isolate, Inverted Reset。

Calibrating Pipettes P i p e t t e C a l i b r a t i o n2P i p e t t e C a l i b r a t i o nEfficient Calibration of Pipettes for Fast & Traceable ResultsWhether you’re an accredited in-house lab or service provider, pipettecalibrations must be performed accurately, efficiently and must fulfill regu-latory guidelines such as ISO8655.METTLER TOLEDO offers comprehensive solutions for pipette calibration.Our reputation for weighing, pipettes and process-oriented software, combined with our expertise of over 20 METTLER TOLDEO calibration labs worldwide results in a user-oriented, efficient and compliant calibration portfolio like no other manufacturer.of the pipette after calibration and adjustment .Calibration Cert ific ateIssue a calibration cert ificate as proof that the pipette has been tested and adjusted.The Calibration ProcessMETTLER TOLEDO's weighing reputation combined with process oriented software, results in an efficient, reliable and compliant pipette calibration process according to ISO8655.3/pipcalISO 8655 compliantISO 17025 accredited labs cali-brate according to ISO8655 for volumetric devices. Our solutions are fully compliant from the balance for measurements to software, with reports for a fast and secure process.4Fast Micro Pipette Calibration Down to 1 μl VolumesCalibration of smallest volumes down to 1 μl are a challenge to do accu -rately and efficiently. The XPE26PC was designed by specialists and is the benchmark in many calibration laboratories world-wide.The micro balance weighing cell, combined with the motion sensor lid and large evaporation trap enable a focused workflow with accurate results right first time. Additional functions such as the StatusLight, the large waste container and optional sensors support an efficient calib-ration process with trusted results.P i p e t t e C a l i b r a t i o nFast & efficient Continuous calibration Unique durable designThe motion sensor lid and large 80 ml evaporation trap of the XPE26PC ensure fewer handling steps and fast, stable results even when calibrating micro pipettes down to 1 μl according to ISO 8655.The large 10 ml container allows many calibrations in a largerange from 1 μl or higher without interruption. A suction pump is included to easily empty the waste water container.Thanks to the unique hanging weighing pan design, the bottom plate is completely sealed and prevents water or fluids dripping into the balance mechanics and weighing cell, for a durable andconsistent lifetime performance.5/pipcalValid results Improve ergonomics Stability increases speedThe built-in StatusLight™ uses color to indicate intuitively the status of the balance. The green light indicates clearly that the balance is ready for you to begin your next calibration, so you can be sure your results will be valid.Optimize your process with an optional motion sensor which can be placed freely on the bench for touchless and conta-mination-free operation of the balance.An optional dual framed weig-hing table reduces to a minimum any user or environmental vibra-tions, substantially speeding up measurement times. With this complete solution, pipette calib-ration is faster than ever.6Calibrate from 20 μl to 10 ml on One BalanceThe XPE analytical balances with professional evaporation traps make it possible to calibrate volumes from 20 μl to 10 ml with just one balance, according to ISO8655 quickly and efficiently.In addition, the unique design with hanging pan and detachable draft shield facilitate optimal working ergonomics and easy cleaning, without the risk of fluids or dirt entering the weighing cell.P i p e t t e C a l i b r a t i o nFrom 2 μl to 2000 μl From 2 ml to 10 ml Simply adapt to your needsGlass evaporation traps enable fast calibrations from 20 microli-ters to 2 milliliters, thanks to the stable setup, good visibility and independent waste container.Simply switch the trap to cali-brate volumes from 2 to 10milliliters. The metal 100 milliliter evaporation container is easily exchanged on the XPE analytical balance.The terminal, motion sensor doors and draft shield can simply be detached anytime to improve access and ergonomicsor just to make cleaning easier.7/pipcalUnique construction Easy Leveling Training DayThanks to the unique hanging waste container, the base is completely sealed which pre-vents dust, fluids and powder from entering the weighing cell ensuring a lasting performance of your balance.The LevelGuide™ provides you with a warning when the XPE balance is not leveled. Full inst-ructions and a graphical leveling bubble are shown on the touch-screen so you can proceed your calibration within e the built-in training module and Pipette Check to train users or check pipettes independently of the system. Users are guided through the process and their individual performance can betracked.8Efficient and Compliant Calibrations with Calibry SoftwareCalibration task planning, guided calibration processes, collecting data, generating reports and being compliant at the same time, these are the benefits of using Calibry Software – effiency personified.Calibry was developed with calibration professionals, who know what is important and necessary in accredited calibration laboratories. With over decades of experience and continuous improvements, Calibry is an established benchmark for many calibration laboratories all over the world.P i p e t t e C a l i b r a t i o nGuided process Complaint and traceable Clever SolutionsCalibry software supports all steps of the calibration process. The large cockpit screen during calibration ensures an easy and secure calibration process of all pipettes.Calibry software saves all calib-ration data in a secure SQL data-base, and ensures compliance with ISO8655 regulations and reports.RFID Pipettes from Rainin need only to be scanned to initiate a task overview or to start a calib-ration directly.9Customized reports 21CFR Part 11 supportive Software validationISO8655 compliant and compre-hensive reports can be generated and adapted to specific needs. If required all data can be exported and used to generate custom specific reports as well.For regulated areas Calibry contains many FDA 21 CFR Part 11supportive features, such as user management, report release, method change history and audit trail.The software validation binder saves valueable time for prepa-ration and validation process. Due to the prefilled information and templates, software releases are completed quickly and more accurately./pipcal10Pipette Calibration Systems Technical InformationCalibration SystemXPE26PCCalibration SystemXPE205Calibration range according ISO8655≥ 1µL – 1'000µL > 10 µL – 10'000 µLBalance type XPE26PC XPE205Evaporation trapBuilt-in 80 mlOptional: 20 ml - 11140043Optional:100 ml - 11138440Motion Sensor Lid opening •-Motion Sensor Draft Shield doors-•Hanging Pan design with closed bottom plate for safe cleaning ••Detachable draft shield -•Detachable terminal ••LevelControl ••StatusLight••ProFact internal Temp. & Time adjustment ••Extra Motion Sensor and/or Footswitch Light Barrier - 11140029ErgoSens - 11132601Footswitch - 11106741Footswitch - 11106741Waste water suction pump •Optional 11138268CarePac Test Weight Set Optional - 11123006Optional - 11123008Calibry Software compatability ••Weighing tableOptional - 11138041Optional - 11138042• included / - not applicableBuilt your own Calibration System that suits your needs.A Calibration System is based on an XPE balance with optional accessories to improve your calibration process. Every balance can be connected to Calibry Software for secure data capture.T e c h n i c a l S p e c i f i c a t i o ns11Model XPE26PC XPE205Order no.3010590130087653Typical Pipette Calibration Range* 1 - 200 µl 20 - 10'000 µl Weighing range 0 - 22 g 0 - 220 g Readability 0.001 mg 0.01 mg Repeatability 0.0015 mg 0.015 mg*) according ISO8655Balance AccessoriesItem Order no.Item Order No.RFID Reader for Calibry Software 30215407Motion Sensor typ "Light Barrier"11140029USB-RS232 Cable 30091827FootSwitch 11106741Ethernet Interface 11132515Weighing Table XPE26PC 11138041RS232 Interface 11132500Weighing Table XPE20511138042Suction pump with tubes 11138268CarePac S Testweight set 1g / 20g 11123006Motion Sensor typ ErgoSens 11132601CarePac S Testweight set 10g / 200g 11123001Pipette Data Management Software functionalities Calibry Single Station Calibry NetworkSoftware package, 1 balance license included 1113841911138420Windows 7 (SP1), 8 and 10 compatible ••Number of max.balances supported 520Databases to manage Pipettes, Methods and Customers ••Calibration Task Planning with reminder ••Calibration report with customization ••ISO8655 compliant and free definable Methods ••Calibration Methods with split, As Found, Maintenance, As Return data••Pipette History data and trending of performance ••Z-factor according to ISO8655, TR or custom with formula editor••Advanced settings for additional uncertainty calculations ••RFID Calibration & Service tracking with MethodCards, SmartTags and Rainin pipettes••System Testing with balance internal & test weights ••Calibry License OptionsInstrument License 1 - to connect 1 additional balance 3041576830415768Instrument License 3 - to connect 3 additional balance 3041732630417326Instrument License 5 - to connect 5 additional balances -30417327Instrument License 1 - to connect 1 other brand balance 3042138230421382User Management License - to manage user rights 30415770included Audit Trail License - to document all system changes 3041577130415771Export File license - - to export in XML and MS Office 30417468included Capture Tool License - to automatically integrate allenviromental data 3041576930415769Evaporation trapsModel Evaporation Trap 10ml Evaporation Trap 20 ml Evaporation Trap 100 ml Order no.111400411114004311138440Compatible with XPE 26 / 56 Micro balance XS / XPE Analytical balance XS /XPE Analytical balance Typical Calibration Range 1 - 200 µl 20 - 1'000 µl 100 - 10'000 ul Volume container 10 ml 6 ml & 20 ml 100 ml Material Aluminium / POM-esd Aluminium / GlassAluminium / POM-esdMETTLER TOLEDO GroupLaboratory DivisionLocal contact: /contactsSubject to technical changes© 10/2017 METTLER-TOLEDO. All rights reserved30393683Global MarCom 2280 LK/PH /pipcalExpect more from usWith decades of experience in pipetting & weighing METTLER TOLEDO can offer you a wide range of online learning resources. Take advantage of our expertise to enhance your know-how and make the most out of your equipment. Check out the services on our internet for a wide range of relevant services and professional solutions we offer.Installation and QualificationTake the stress-free route to meet your producti-vity, quality and regulatory requirements imme-diately. Your specific needs can be addressedby choosing the most suitable option from our comprehensive service offering./serviceSmartStand – Never miss a calibrationAvoid costly efforts and retry's with out-of-specification pipettes or those are beyond cali-bration date. Rainin XLS+ with RFID tags showtheir status every time you put them on the stand.Combined with the EasyDirect Software you can monitor and manage all pipettes in your lab./smartstandRainin pipettes – performance you can trustIts comfortable handle, light springs and “Magne-tic Assist™” technology, ensure light and smoothoperation, while significantly reducing the risk ofrepetitive strain injuries. Tip shaft options includedlow-force LTS™ for improved ergonomics and universal fit./rainin。

无损检测中英文对照无损检测常用词汇中英文对照表English 中文释义A.C magnetic saturation 交流磁饱和Absorbed dose 吸收剂量Absorbed dose rate 吸收剂量率Acceptance criteria验收准则Acceptanc limits 验收范围Acceptable quality level可验收质量等级Acceptance level 验收水平（验收等级）Acceptance standard 验收标准Accumulation test 累积检测Acoustic emission count（emission count）声发射计数（发射计数）Acoustic emission transducer 声发射换能器（声发射传感器）Acoustic emission(AE) 声发射Acoustic holography 声全息术Acoustic impedance 声阻抗Acoustic impedance matching 声阻抗匹配Acoustic impedance method 声阻法Acoustic wave 声波Acoustical lens 声透镜Acoustic—ultrasonic 声-超声（AU）Activation 活化Activity 活度Adequate shielding 安全屏蔽Ampere turns 安匝数Amplitude 幅度Angle beam method 斜射法Angle of incidence 入射角Angle of reflection 反射角Angle of spread 指向角Angle of squint 偏向角Angle probe 斜探头Angstrom unit 埃(A)Area amplitude response curve 面积幅度曲线 Area of interest 评定区Artificial discontinuity 人工不连续性Artifact 假缺陷Artificial defect 人工缺陷Artificial discontinuity 标准人工缺陷A-scan A 型扫描A-scope; A-scan A 型显示Attenuation coefficient 衰减系数Attenuator 衰减器Audible leak indicator 音响泄漏指示器Automatic testing 自动检测 Autoradiography 自射线照片Avaluation 评定Background背景Barium concrete 钡混凝土Barn 靶Base fog 片基灰雾Bath 槽液（浸湿）Bayard- Alpert ionization gage B- A 型电离计Beam 声束Beam ratio 光束比Beam angle 束张角Beam axis 声束轴线Beam index 声束入射点Beam path location 声程定位Beam path; path length 声程Beam spread 声束扩散Betatron 电子感应加速器Bimetallic strip gage 双金属片计Bipolar field 双极磁场Black light filter 黑光滤波器Black light; ultraviolet radiation 黑光Blackbody 黑体Blackbody equivalent temperature 黑体等效温度Bleakney mass spectrometer 波利克尼质谱仪 Bleedout 渗出Bottom echo 底面回波Bottom surface 底面Boundary echo(first) 边界一次回波Bremsstrahlung 轫致辐射Broad-beam condition 宽射束Brush application 刷涂B-scan presenfation B 型扫描显示B-scope; B-scan B 型显示C- scan C 型扫描Calibration, instrument 设备校准Capillary action 毛细管作用Carrier fluid 载液Carry over of penetrate 渗透剂移转Cassette 暗合Cathode 阴极Central conductor method 中心导体法Characteristic curve 特性曲线Characteristic curve of film 胶片特性曲线Characteristic radiation 特征辐射Chemical fog 化学灰雾Cine-radiography 射线(活动)电影摄影术Cintact pads 接触垫Circumferential coils 圆环线圈Circumferential field 周向磁场Circumferential magnetization method 周向磁化法Clean 清理Clean- up 清除Clearing time 定透时间Coercive force 矫顽力Coherence 相干性Coherence length 相干长度（谐波列长度） Coi1，test 测试线圈Coil size 线圈大小Coil spacing 线圈间距Coil technique 线圈技术Coil method 线圈法Coilreference 线圈参考Coincidence discrimination 符合鉴别Cold-cathode ionization gage 冷阴极电离计 Collimator 准直器Collimation 准直Collimator 准直器Colour contrast penetrant着色渗透剂Combined colour comtrast and fluorescent pe netrant 着色荧光渗透剂Compressed air drying 压缩空气干燥Compressional wave 压缩波Compton scatter 康普顿散射Continuous emission 连续发射Continuous linear array 连续线阵Continuous method 连续法Continuous spectrum 连续谱Continuous wave 连续波Contract stretch 对比度宽限Contrast 对比度Contrast agent 对比剂Contrast aid 反差剂Contrast sensitivity 对比灵敏度Control echo 监视回波Control echo 参考回波Couplant 耦合剂Coupling 耦合Coupling losses 耦合损失Cracking 裂解Creeping wave 爬波Critical angle 临界角Cross section 横截面Cross talk 串音 Cross-drilled hole 横孔Crystal 晶片C-scope; C-scan C 型显示Curie point 居里点Curie temperature 居里温度Curie(Ci) 居里Current flow method 通电法Current induction method 电流感应法Current magnetization method 电流磁化法Cut－off level 截止电平Dead zone 盲区Decay curve 衰变曲线Decibel(dB) 分贝Defect 缺陷Defect resolution 缺陷分辨力Defect detection sensitivity 缺陷检出灵敏度Detection sensitivity检测灵敏度Detection threshold检测门槛Defect resolution 缺陷分辨力Definition 清晰度Definition， image definition 清晰度，图像清晰度Demagnetization 退磁Demagnetization factor 退磁因子Demagnetizer 退磁装置Densitometer 黑度计Density 黑度（底片）Density comparison strip 黑度比较片Detecting medium 检验介质Detergent remover 洗净液Developer 显像剂Developer， agueons 水性显象剂Developer， dry 干显象剂Developer， liquid film 液膜显象剂Developer， nonaqueous （sus- pendible）非水（可悬浮）显象剂Developing time 显像时间Development 显影Diffraction mottle 衍射斑Diffuse indications 松散指示Diffusion 扩散Digital image acquisition system 数字图像识别系统Dilatational wave 膨胀波Dip and drain station 浸渍和流滴工位Dip rinse浸洗Direct contact magnetization 直接接触磁化Direct exposure imaging 直接曝光成像Direct contact method 直接接触法Directivity 指向性Discontinuity 不连续性Distance- gain- size-German A VG 距离- 增益- 尺寸（DGS 德文为A VG）Distance marker; time marker 距离刻度Dose equivalent 剂量当量Dose rate meter 剂量率计Dosemeter 剂量计Double crystal probe 双晶片探头Double probe technique 双探头法Double transceiver technique 双发双收法Double traverse technique 二次波法Dragout 带出Drain time 滴落时间Drain time 流滴时间Drift 漂移Dry method 干法Dry powder 干粉Dry technique 干粉技术Dry developer 干显显像剂Dry developing cabinet 干显像柜Dry method 干粉法Drying oven 干燥箱Drying station 干燥工位Drying time 干燥时间D-scope; D-scan D 型显示Dual-focus tube 双焦点管Dual purpose penetrant 两用渗透剂Dual search unit 双探头Duplex-wire image quality indicator 双线像质指示器Duration 持续时间Dwell time 停留时间Dye penetrant 着色渗透剂Dynamic leak test 动态泄漏检测Dynamic leakage measurement 动态泄漏测量Dynamic range 动态范围Dynamic radiography 动态射线透照术Echo 回波Echo frequency 回波频率Echo height 回波高度Echo indication 回波指示Echo transmittance of sound pressure 往复透过率Echo width 回波宽度Eddy current 涡流 Eddy current flaw detector 涡流探伤仪Eddy current testiog 涡流检测Edge 端面Edge effect 边缘效应Edge echo 棱边回波Edge effect 边缘效应Effective depth penetration （EDP）有效穿透深度Effective focus size 有效焦点尺寸Effective magnetic permeability 有效磁导率Effective permeability 有效磁导率Effective reflection surface of flaw 缺陷有效反射面Effective resistance 有效电阻Elastic medium 弹性介质Electric displacement 电位移Electrostatic spraying 静电喷射Electrical center 电中心Electrode 电极Electromagnet 电磁铁Electro-magnetic acoustic transducer 电磁声换能器Electromagnetic induction 电磁感应Electromagnetic radiation 电磁辐射Electromagnetic testing 电磁检测Electro-mechanical coupling factor 机电耦合系数Electron radiography 电子辐射照相术Electron volt 电子伏恃Electronic noise 电子噪声Electrostatic spraying 静电喷涂Emulsification 乳化Emulsification of penetrant 渗透剂的乳化 Emulsification time 乳化时间Emulsifier 乳化剂Encircling coils 环绕式线圈End effect 端部效应Energizing cycle 激励周期Equalizing filter 均衡滤波器Equivalent 当量Equivalent I．Q. I. Sensitivity 象质指示器当量灵敏度Equivalent nitrogen pressure 等效氮压Equivalent penetrameter sensifivty 透度计当量灵敏度Equivalent method 当量法Erasabl optical medium 可探光学介质Etching 浸蚀Evaluation 评定Evaluation threshold 评价阈值Event count 事件计数Event count rate 事件计数率Examination area 检测范围Examination region 检验区域Excess penetrant removal多余渗透剂的去除Exhaust pressure/discharge pressure 排气压力Exhaust tubulation 排气管道Expanded time-base sweep 时基线展宽Exposure 曝光Exposure table 曝光表格Exposure chart 曝光曲线Exposure fog 曝光灰雾Exposure，radiographic exposure 曝光，射线照相曝光Extended source 扩展源Facility scattered neutrons 条件散射中子False indication 假指示（伪显示）Family 族Far field 远场Feed-through coil 穿过式线圈Field， resultant magnetic 复合磁场Fill factor 填充系数Film speed 胶片速度Film badge 胶片襟章剂量计Film base 片基Film contrast 胶片对比度Film gamma 胶片γ值Film processing 胶片冲洗加工Film speed 胶片感光度Film unsharpness 胶片不清晰度Film viewing screen 观察屏Filter 滤波器/滤光板Final test 复探Flat-bottomed hole 平底孔Flat-bottomed hole equivalent 平底孔当量Flaw 伤Flaw characterization 伤特性Flaw echo 缺陷回波Flexural wave 弯曲波Floating threshold 浮动阀值Fluorescence 荧光Fluorescent examination method 荧光检验法Fluorescent intensity 荧光强度Fluorescent penetrant 荧光渗透剂Fluorescent magnetic particle inspection 荧光磁粉检验Fluorescent dry deposit penetrant 干沉积荧光渗透剂Fluorescent light 荧光Fluorescent magnetic powder 荧光磁粉Fluorescent penetrant 荧光渗透剂Fluorescent screen 荧光屏Fluoroscopy 荧光检查法Flux leakage field 磁通泄漏场Flux lines 磁通线Focal spot 焦点Focal distance 焦距Focus length 焦点长度Focus size 焦点尺寸Focus width 焦点宽度Focus(electron) 电子焦点Focused beam 聚焦声束Focusing probe 聚焦探头Focus-to-film distance(f.f.d) 焦点-胶片距离（焦距）Fog 底片灰雾Fog density 灰雾密度Footcandle 英尺烛光Freguency 频率Frequency constant 频率常数Fringe 干涉带Front distance 前沿距离Front distance of flaw 缺陷前沿距离Full- wave direct current（FWDC）全波直流Fundamental frequency 基频Furring 毛状迹痕Gage pressure 表压Gain 增益Gamma radiography γ射线透照术Gamma ray source γ射线源Gamma ray source container γ射线源容器Gamma rays γ射线Gamma-ray radiographic equipment γ射线透照装置Gap scanning 间隙扫查Gas 气体Gate 闸门Gating technique 选通技术Gauss 高斯Geiger-Muller counter 盖革.弥勒计数器Geometric unsharpness 几何不清晰度Gray(Gy) 戈瑞Grazing incidence 掠入射Grazing angle 掠射角Group velocity 群速度Half life 半衰期Half- wave current （HW）半波电流Half-value layer(HVL) 半值层Half-value method 半波高度法Halogen 卤素Halogen leak detector 卤素检漏仪Hard X-rays 硬X 射线Hard-faced probe 硬膜探头Harmonic analysis 谐波分析Harmonic distortion 谐波畸变Harmonics 谐频Head wave 头波Helium bombing 氦轰击法Helium drift 氦漂移Helium leak detector 氦检漏仪Hermetically tight seal 气密密封High vacuum 高真空High energy X-rays 高能X 射线Holography (optical) 光全息照相Holography， acoustic 声全息Hydrophilic emulsifier 亲水性乳化剂Hydrophilic remover 亲水性洗净剂Lipophilic emulsifier 亲油性乳化剂Hydrostatic text 流体静力检测Hysteresis 磁滞Hysteresis 磁滞IACS IACSID coil ID 线圈Image definition 图像清晰度Image contrast 图像对比度Image enhancement 图像增强Image magnification 图像放大Image quality 图像质量Image quality indicator sensitivity 像质指示器灵敏度Image quality indicator(IQI)/image quality ind ication 像质指示器Imaging line scanner 图像线扫描器Immersion probe 液浸探头Immersion rinse 浸没清洗Immersion testing 液浸法Immersion time 浸没时间Impedance 阻抗Impedance plane diagram 阻抗平面图Imperfection 不完整性（缺欠）Impulse eddy current testing 脉冲涡流检测Incremental permeability 增量磁导率Indicated defect area 缺陷指示面积Indicated defect length 缺陷指示长度Indication 指示（显示）Indirect exposure 间接曝光Indirect magnetization 间接磁化Indirect magnetization method 间接磁化法 Indirect scan 间接扫查Induced field 感应磁场Induced current method 感应电流法Infrared imaging system 红外成象系统Infrared sensing device 红外扫描器Inherent fluorescence 固有荧光Inherent filtration 固有滤波Initial permeability 起始磁导率Initial pulse 始脉冲Initial pulse width 始波宽度Inserted coil 插入式线圈Inside coil 内部线圈Inside- out testing 外泄检测Inspection 检查Inspection medium 检查介质Inspection frequency/ test frequency 检测频率Intensifying factor 增感系数Intensifying screen 增感屏Interal,arrival time （Δtij)/arrival time interval （Δtij）到达时间差(Δtij)Interface boundary 界面Interface echo 界面回波Interface trigger 界面触发Interference 干涉Interpretation 解释Ion pump 离子泵Ion source 离子源Ionization chamber 电离室Ionization potential 电离电位Ionization vacuum gage 电离真空计Ionography 电离射线透照术Irradiance， E 辐射通量密度， EIsolation 隔离检测Isotope 同位素Kaiser effect 凯塞(Kaiser)效应Kilo volt kv 千伏特Kiloelectron volt keV 千电子伏特Krypton 85 氪85L／D ratio L/D 比Lamb wave 兰姆波Latent image 潜象Lateral scan 左右扫查Lateral scan with oblique angle 斜平行扫查Latitude (of an emulsion) 胶片宽容度Lead screen 铅屏Leak 泄漏孔Leak artifact 泄漏器Leak detector 检漏仪Leak testtion 泄漏检测Leakage field 泄漏磁场Leakage rate 泄漏率Leechs 磁吸盘Lift-off effect 提离效应Light intensity 光强度Limiting resolution 极限分辨率Line scanner 线扫描器Line focus 线焦点Line pair pattern 线对检测图Line pairs per millimetre 每毫米线对数Linear (electron) accelerator(LINAC) 电子直线加速器Linear attenuation coefficient 线衰减系数Linear scan 线扫查Linearity （time or distance）线性（时间或距离）Linearity， anplitude 幅度线性Lines of force 磁力线Lipophilic emulsifier 亲油性乳化剂Lipophilic remover 亲油性洗净剂Liquid penetrant examination 液体渗透检验Liquid filmdeveloper 液膜显像剂Local magnetization 局部磁化Local magnetization method 局部磁化法Local scan 局部扫查Localizing cone 定域喇叭筒Location 定位Location accuracy 定位精度Location computed 定位，计算Location marker 定位标记Location upon delta-T 时差定位Location， clusfer 定位，群集Location，continuous AE signal 定位，连续AE 信号Longitudinal field 纵向磁场Longitudinal magnetization method 纵向磁化法Longitudinal resolution 纵向分辨率Longitudinal wave 纵波 Longitudinal wave probe 纵波探头Longitudinal wave technique 纵波法Loss of back reflection 背面反射损失Loss of back reflection 底面反射损失Love wave 乐甫波Low energy gamma radiation 低能γ辐射Low－enerugy photon radiation 低能光子辐射Luminance 亮度Luminosity 流明Lusec 流西克Maga or million electron volts MeV 兆电子伏特Magnetic history 磁化史Magnetic hysteresis 磁性滞后Magnetic particle field indication 磁粉磁场指示器Magnetic particle inspection flaw indications 磁粉检验的伤显示Magnetic circuit 磁路Magnetic domain 磁畴Magnetic field distribution 磁场分布Magnetic field indicator 磁场指示器Magnetic field meter 磁场计Magnetic field strength 磁场强度(H)Magnetic field/field，magnetic 磁场Magnetic flux 磁通Magnetic flux density 磁通密度Magnetic force 磁化力Magnetic leakage field 漏磁场Magnetic leakage flux 漏磁通Magnetic moment 磁矩Magnetic particle 磁粉Magnetic particle indication 磁痕Magnetic particle testing/magnetic particle ex amination 磁粉检测Magnetic permeability 磁导率Magnetic permeability 磁导率Magnetic pole 磁极Magnetic saturataion 磁饱和Magnetic saturation 磁饱和Magnetic writing 磁写Magnetizing 磁化Magnetizing current 磁化电流Magnetizing coil 磁化线圈Magnetostrictive effect 磁致伸缩效应Magnetostrictive transducer 磁致伸缩换能器 Main beam 主声束Manual testing 手动检测Markers 时标MA-scope; MA-scan MA 型显示Masking 遮蔽Mass attcnuation coefficient 质量吸收系数 Mass number 质量数Mass spectrometer （M.S.）质谱仪Mass spectrometer leak detector 质谱检漏仪 Mass spectrum 质谱Master/slave discrimination 主从鉴别MDTD 最小可测温度差Mean free path 平均自由程Medium vacuum 中真空Mega or million volt MV 兆伏特Micro focus X - ray tube 微焦点X 光管Microfocus radiography 微焦点射线透照术 Micrometre 微米Micron of mercury 微米汞柱Microtron 电子回旋加速器Milliampere 毫安（mA）Millimetre of mercury 毫米汞柱Minifocus x- ray tube 小焦点调射线管Minimum detectable leakage rate 最小可探泄漏率Minimum resolvable temperature difference （MRTD）最小可分辨温度差（MRDT）Mode 波型Mode conversion 波型转换Mode transformation 波型转换Moderator 慢化器Modulation transfer function （MTF）调制转换功能（MTF）Modulation analysis 调制分析Molecular flow 分子流Molecular leak 分子泄漏Monitor 监控器Monochromatic 单色波Movement unsharpness 移动不清晰度Moving beam radiography 可动射束射线透照术Multiaspect magnetization method 多向磁化法Multidirectional magnetization 多向磁化Multifrequency eddy current testiog 多频涡流检测Multiple back reflections 多次背面反射Multiple reflections 多次反射Multiple back reflections 多次底面反射 Multiple echo method 多次反射法Multiple probe technique 多探头法Multiple triangular array 多三角形阵列Narrow beam condition 窄射束NC NCNear field 近场Near field length 近场长度Near surface defect 近表面缺陷Net density 净黑度Net density 净(光学)密度Neutron 中子Neutron radiograhy 中子射线透照Neutron radiography 中子射线透照术Newton （N）牛顿Nier mass spectrometer 尼尔质谱仪Noise 噪声Noise equivalent temperature difference （NE TD）噪声当量温度差（NETD）Nominal angle 标称角度Nominal frequency 标称频率Non-aqueous liquid developer 非水性液体显像剂Non-condensable gas 非冷凝气体Non destructive Examination（NDE）无损试验Non-destructive Examination损检查Nondestructive Evaluation（NDE）无损评价Nondestructive Inspection（NDI）无损检验Nondestructive Testing（NDT）无损检测Nonerasble optical data 可固定光学数据Nonferromugnetic material 非铁磁性材料Non-relevant indication 非相关指示（显示）Non-screen-type film 非增感型胶片Normal incidence 垂直入射（亦见直射声束）Normal permeability 标准磁导率Normal beam method; straight beam method 垂直法Normal probe 直探头Normalized reactance 归一化电抗Normalized resistance 归一化电阻Nuclear activity 核活性Nuclide 核素Object plane resolution 物体平面分辨率Object scattered neutrons 物体散射中子Object beam 物体光束Object beam angle 物体光束角Object-film distance 被检体-胶片距离Object 一film distance 物体- 胶片距离Over development 显影过度Over emulsfication 过乳化Overall magnetization 整体磁化Overload recovery time 过载恢复时间Overwashing 过洗Oxidation fog 氧化灰雾P PPair production 偶生成Pair production 电子对产生Pair production 电子偶的产生Palladium barrier leak detector 钯屏检漏仪Panoramic exposure 全景曝光Parallel scan 平行扫查Paramagnetic material 顺磁性材料Parasitic echo 干扰回波Partial pressure 分压Particle content 磁悬液浓度Particle velocity 质点(振动)速度Pascal （Pa）帕斯卡（帕）Pascal cubic metres per second 帕立方米每秒（Pa?m3/s ）Path length 光程长Path length difference 光程长度差Pattern 探伤图形Peak current 峰值电流Peelable developer 可剥离显像剂Penetrameter 透度计Penetrant 渗透剂Penetrameter sensitivity 透度计灵敏度Penetrant 渗透剂Penetrant comparator 渗透对比试块Penetrant flaw detection 渗透探伤Penetrant removal 渗透剂去除Penetrant station 渗透工位Penetrant，water- washable 水洗型渗透剂Penetration 穿透深度Penetrant testing 渗透检测Penetration time 渗透时间Penetrant materials渗透材料Permanent magnet 永久磁铁Permeability coefficient 透气系数Permeability，a-c 交流磁导率Permeability，d－c 直流磁导率Phantom echo 幻象回波Phase analysis 相位分析Phase angle 相位角Phase controlled circuit breaker 断电相位控制器Phase detection 相位检测Phase hologram 相位全息Phase sensitive detector 相敏检波器Phase shift 相位移Phase velocity 相速度Phase-sensitive system 相敏系统Phillips ionization gage 菲利浦电离计Phosphor 荧光物质Photo fluorography 荧光照相术Photoelectric absorption 光电吸收Photographic emulsion 照相乳剂Photographic fog 照相灰雾Photostimulable luminescence 光敏发光Piezoelectric effect 压电效应Piezoelectric material 压电材料Piezoelectric stiffness constant 压电劲度常数Piezoelectric stress constant 压电应力常数 Piezoelectric transducer 压电换能器Piezoelectric voltage constant 压电电压常数 Pirani gage 皮拉尼计Pirani gage 皮拉尼计Pitch and catch technique 一发一收法Pixel 象素Pixel size 象素尺寸Pixel， disply size 象素显示尺寸Planar array 平面阵(列)Plane wave 平面波Plate wave 板波Plate wave technique 板波法Point source 点源Post emulsification 后乳化Post emulsifiable penetrant 后乳化渗透剂Post-cleaning 后清洗Powder 粉未Powder blower 喷粉器Powder blower 磁粉喷枪Pre-cleaning 预清理Pressure difference 压力差Pressure dye test 压力着色检测Pressure probe 压力探头Pressure testing 压力检测Pressure- evacuation test 压力抽空检测Pressure mark 压痕Pressure,design 设计压力Pre-test 初探Primary coil 一次线圈Primary radiation 初级辐射Probe gas 探头气体Probe test 探头检测Probe backing 探头背衬Probe coil 点式线圈Probe coil 探头式线圈Probe coil clearance 探头线圈间隙Probe index 探头入射点Probe to weld distance 探头-焊缝距离Probe/ search unit 探头Process control radiograph 工艺过程控制的射线照相Processing capacity 处理能力Processing speed 处理速度Prods 触头Projective radiography 投影射线透照术Proportioning probe 比例探头Protective material 防护材料Proton radiography 质子射线透照Product family 产品族Pulse 脉冲波Pulse 脉冲Pulse echo method 脉冲回波法Pulse repetition rate 脉冲重复率Pulse amplitude 脉冲幅度Pulse echo method 脉冲反射法Pulse energy 脉冲能量Pulse envelope 脉冲包络Pulse length 脉冲长度Pulse repetition frequency 脉冲重复频率Pulse tuning 脉冲调谐Pump- out tubulation 抽气管道Pump-down time 抽气时间Q factor Q 值Quadruple traverse technique 四次波法Quality (of a beam of radiation) 射线束的质 Quality factor 品质因数Quenching 阻塞Quenching of fluorescence 荧光的猝灭Quick break 快速断间Rad(rad) 拉德Radiance， L 面辐射率，LRadiant existence， M 幅射照度MRadiant flux； radiant power，ψe辐射通量、辐射功率、ψeRadiation 辐射Radiation does 辐射剂量Radio frequency （r- f） display 射频显示Radio- frequency mass spectrometer 射频质谱仪Radio frequency(r-f) display 射频显示Radiograph 射线底片Radiographic contrast 射线照片对比度Radiographic equivalence factor 射线照相等效系数Radiographic exposure 射线照相曝光量Radiographic inspection 射线检测Radiographic inspection 射线照相检验Radiographic quality 射线照相质量Radiographic sensitivity 射线照相灵敏度Radiographic contrast 射线底片对比度Radiographic equivalence factor 射线透照等效因子Radiographic inspection 射线透照检查Radiographic quality 射线透照质量Radiographic sensitivity 射线透照灵敏度Radiography 射线照相术Radiological examination 射线检验Radiology 射线学Radiometer 辐射计Radiometry 辐射测量术Radioscopy 射线检查法Range 量程Rayleigh wave 瑞利波Rayleigh scattering 瑞利散射Real image 实时图像Real-time radioscopy 实时射线检查法Rearm delay time 重新准备延时时间Rearm delay time 重新进入工作状态延迟时间Reciprocity failure 倒易律失效Reciprocity law 倒易律Recording medium 记录介质Recovery time 恢复时间Rectified alternating current 脉动直流电Reference block 参考试块Reference beam 参考光束Reference block 对比试块Reference block method 对比试块法Reference coil 参考线圈Reference line method 基准线法Reference standard 参考标准Reference test piece参考试件Reflection 反射Reflection coefficient 反射系数Reflection density 反射密度Reflector 反射体Refraction 折射Refractive index 折射率Refrence beam angle 参考光束角Reicnlbation 网纹Reject; suppression 抑制Rejection level 拒收水平Relative permeability 相对磁导率Relevant indication 相关指示（显示）Reluctance 磁阻Rem(rem) 雷姆Remote controlled testing 机械化检测Replenisers 补充剂Representative quality indicator 代表性质量指示器Residual magnetic field/field， residual magn etic 剩磁场Residual technique 剩磁技术Residual magnetic method 剩磁法Residual magnetism 剩磁Resistance （to flow）气阻Resolution 分辨力Resonance method 共振法Response factor 响应系数Response time 响应时间Resultant field 复合磁场Resultant magnetic field 合成磁场Resultant magnetization method 组合磁化法 Retentivity 顽磁性Reversal 反转现象Ring-down count 振铃计数Ring-down count rate 振铃计数率Rinse 清洗（冲洗）Rise time 上升时间Rise-time discrimination 上升时间鉴别Rod-anode tube 棒阳极管Roentgen(R) 伦琴Roof angle 屋顶角Rotational magnetic field 旋转磁场Rotational magnetic field method 旋转磁场法 Rotational scan 转动扫查Roughing 低真空Roughing line 低真空管道Roughing pump 低真空泵S SSafelight 安全灯Sampling probe 取样探头Saturation 饱和Saturation，magnetic 磁饱和Saturation level 饱和电平 Scan on grid lines 格子线扫查Scan pitch 扫查间距Scanning 扫查Scanning index 扫查标记Scanning directly on the weld 焊缝上扫查Scanning path 扫查轨迹Scanning sensitivity 扫查灵敏度Scanning speed 扫查速度Scanning zone 扫查区域Scattared energy 散射能量Scatter unsharpness 散射不清晰度Scattered neutrons 散射中子Scattered radiation 散射辐射Scattering 散射Schlieren system 施利伦系统Scintillation counter 闪烁计数器Scintillator and scintillating crystals 闪烁器和闪烁晶体Screen 屏Screen unsharpness 荧光增感屏不清晰度Screen-type film 荧光增感型胶片SE probe SE 探头Search-gas 探测气体Second critical angle 第二临界角Secondary radiation 二次射线Secondary coil 二次线圈Secondary radiation 次级辐射Selectivity 选择性Semi-conductor detector 半导体探测器Sensitirity va1ue 灵敏度值Sensitivity 灵敏度Sensitivity of leak test 泄漏检测灵敏度Sensitivity level (of a penentrant inspection pr ocess) (渗透检测过程的)灵敏度等级Sensitivity control 灵敏度控制Shear wave 切变波Shear wave probe 横波探头Shear wave technique 横波法Shim 薄垫片Shot 冲击通电Side lobe 副瓣Side wall 侧面Sievert(Sv) 希(沃特)Signal 信号Signal gradient 信号梯度Signal over load point 信号过载点Signal overload level 信号过载电平Signal to noise ratio 信噪比Single crystal probe 单晶片探头Single probe technique 单探头法Single traverse technique 一次波法Sizing定量Sizing technique 定量法Skin depth 集肤深度Skin effect 集肤效应Skip distance 跨距Skip point 跨距点Sky shine(air scatter) 空中散射效应Sniffing probe 嗅吸探头Soft X-rays 软X 射线Soft-faced probe 软膜探头Solarization 负感作用Solenoid 螺线管Solvent based developer 溶剂型显像剂Soluble developer 可溶显像剂Solvent remover 溶剂去除剂Solvent-removable penentrant 溶剂去除型渗透剂Solvent cleaners 溶剂清除剂Solvent developer 溶剂显像剂Solvent remover 溶剂洗净剂Solvent-removal penetrant 溶剂去除型渗透剂Sorption 吸着Sound diffraction 声绕射Sound insulating layer 隔声层Sound intensity 声强Sound intensity level 声强级Sound pressure 声压Sound scattering 声散射Sound transparent layer 透声层Sound velocity 声速Source 源Source data label 放射源数据标签Source location 源定位Source size 源尺寸Source-film distance 射线源-胶片距离Spacial frequency 空间频率Spark coil leak detector 电火花线圈检漏仪 Specific activity 放射性比度Specified sensitivity 规定灵敏度Standard 标准Standard 标准试样Standard leak rate 标准泄漏率Standard leak 标准泄漏孔Standard tast block 标准试块 Standardization instrument 设备标准化Standing wave; stationary wave 驻波Step wedge 阶梯楔块Step- wadge calibration film 阶梯楔块校准底片Step- wadge comparison film 阶梯楔块比较底片Step wedge 阶梯楔块Step-wedge calibration film 阶梯-楔块校准片Step-wedge comparison film 阶梯-楔块比较片Stereo-radiography 立体射线透照术Subject contrast 被检体对比度Subsurface discontinuity 近表面不连续性Suppression 抑制Surface echo 表面回波Surface field 表面磁场Surface noise 表面噪声Surface wave 表面波Surface wave probe 表面波探头Surface wave technique 表面波法Surge magnetization 脉动磁化Surplus sensitivity 灵敏度余量Suspension 磁悬液Sweep 扫描Sweep range 扫描范围Sweep speed 扫描速度Swept gain 扫描增益Swivel scan 环绕扫查System exanlillatien threshold 系统检验阈值System inclacel artifacts 系统感生物System noise 系统噪声Tackground， target 目标本底Tandem scan 串列扫查Target 耙Target 靶Testing products 检测产品Television fluoroscopy 电视X 射线荧光检查Temperature envelope 温度范围Tenth-value-layer(TVL) 十分之一值层Test coil 检测线圈Test quality level 检测质量水平Test ring 试环Test block 试块Test frequency 试验频率Test piece 试片Test range 探测范围Test surface 探测面Testing，ulrasonic 超声检测Thermal neutrons 热中子Thermocouple gage 热电偶计Thermogram 热谱图Thermography， infrared 红外热成象Thermoluminescent dosemeter(TLD) 热释光剂量计Thickness sensitivity 厚度灵敏度Third critiical angle 第三临界角Thixotropic penetrant 摇溶渗透剂Thormal resolution 热分辨率Threading bar 穿棒Three way sort 三档分选Threshold setting 门限设置Threshold fog 阈值灰雾Threshold level 阀值Threshotd tcnet 门限电平Throttling 节流Through transmission technique 穿透技术Through penetration technique 贯穿渗透法Through transmission technique; transmission technique 穿透法Through-coil technique 穿过式线圈技术Throughput 通气量Tight 密封Total reflection 全反射Totel image unsharpness 总的图像不清晰度 Tracer probe leak location 示踪探头泄漏定位Tracer gas 示踪气体Transducer 换能器/传感器Transition flow 过渡流Translucent base media 半透明载体介质Transmission 透射Transmission densitomefer 发射密度计Transmission coefficient 透射系数Transmission point 透射点Transmission technique 透射技术Transmittance，τ透射率τTransmitted film density 检测底片黑度Transmitted pulse 发射脉冲Transverse resolution 横向分辨率Transverse wave 横波Traveling echo 游动回波Travering scan; depth scan 前后扫查Triangular array 正三角形阵列Trigger/alarm condition 触发/报警状态 Trigger/alarm level 触发/报警标准Triple traverse technique 三次波法True continuous technique 准确连续法技术Trueattenuation 真实衰减Tube current 管电流Tube head 管头Tube shield 管罩Tube shutter 管子光闸Tube window 管窗Tube-shift radiography 管子移位射线透照术 Two-way sort 两档分选Ultra- high vacuum 超高真空Ultrasonic leak detector 超声波检漏仪Ultrasonic noise level 超声噪声电平Ultrasonic cleaning 超声波清洗Ultrasonic field 超声场Ultrasonic flaw detection 超声探伤Ultrasonic flaw detector 超声探伤仪Ultrasonic microscope 超声显微镜Ultrasonic spectroscopy 超声频谱Ultrasonic testing system 超声检测系统Ultrasonic thickness gauge 超声测厚仪Ultraviolet radiation 紫外辐射Under development 显影不足Unsharpness 不清晰Useful density range 有效光学密度范围UV-A A 类紫外辐射UV-A filter A 类紫外辐射滤片Vacuum 真空Vacuum cassette 真空暗盒Vacuum testing 真空检测Vacuum cassette 真空暗合Van de Graaff generator 范德格喇夫起电机 Vapor pressure 蒸汽压Vapour degreasing 蒸汽除油Variable angle probe 可变角探头Vee path V 型行程Vehicle 载体Vertical linearity 垂直线性Vertical location 垂直定位Viewing 观察Viewing conditions观察条件Visible light 可见光Vitua limage 虚假图像。

Probabilistic Neural Networks inBankruptcy PredictionZ.R.YangU NIVERSITY OF P ORTSMOUTHMarjorie B.PlattN ORTHEASTERN U NIVERSITYHarlan D.PlattN ORTHEASTERN U NIVERSITYPrior application of neural networks in finance and accounting,notably ogy has also demonstrated useful advantages in other financialapplications,including futures trading volumes prediction in bankruptcy prediction,are limited to back-propagation neural networks.Their well-known disadvantages,however,limit the practical usefulness(Kaastra and Boyd,1995),stocks selection(Kryzanowski,Galler,and Wright,1993),forecasting exchange rates(Kuan of neural discriminant models.Instead,a probabilistic neural networkwith fewer of these difficulties is ing data from the U.S.and Liu,1995),and real estate valuation(Worzala,Lenk,andSilva,1995).oil and gas industry,alternate methodologies are compared.Whereasprobabilistic neural networks without pattern normalization and Fisher These studies rely on back-propagation NN techniques asdescribed by Rumelhart and McClelland(1986).Salchen-discriminant analysis achieve the best overall estimation results,discrimi-nant analysis produces superior results for bankrupt companies.J BUSN berger,Cinar,and Lash(1992)document numerous disadvan-tages of the back-propagation method including its computa-RES1999.44.67–74.©1998Elsevier Science Inc.tional intensity,its inability to explain conclusions or howthey are reached,and its lack of formal theory which imposesa need for expertise on the user.In addition,Altman,Marco, B ankruptcy prediction models have used a variety of and Varetto(1994)note that it may yield illogical behavioralstatistical methodologies,resulting in varying out-estimates when input values are varied.comes.These methodologies include linear discrimi-Another approach uses Specht’s(1990)probabilistic NN nant analysis(Altman,1968;Altman,Haldeman,and Naraya-that exhibits both feed-forward(i.e.,the difference between nan,1977),regression analysis(Korobow,Stuhr,and Martin,a model’s targets and outputs are not fed back to modifyparameters)and fast learning,which increases its computa-1976),logit regression(Barth,et al.,1985;Pantalone and Platt,1987),and weighted average maximum likelihood estimation tional speed.Another advantage of probabilistic NNs is how (Zmijewski,1984).The purpose of this study is to directlythey respect the statistician’s mantra to not waste data:back-propagation systems require a validation data set(i.e.,wasted compare linear discriminant analysis,an accepted statisticalcases)to search for over fitting,whereas probabilistic NNs, methodology,with several neural network(NN)techniqueswhich are designed for classification,use all available data in to determine which approach is the most useful in providingmodel building.accurate,reliable bankruptcy predictions among firms in theIn this study,empirical results are compared across dis-oil and gas industry.criminant analysis,back-propagation NNs,and two varieties The possible contribution of NNs to the early warningof probabilistic NNs,using a single data set of oil and gas failure prediction literature has been demonstrated recentlycompanies.The primary finding for the entire sample based for a variety of industries(see Salchenberger,Cinar,and Lash,on deflated financial ratios was that the discriminant analysis 1992;Coats and Fant,1993;Altman,Marco,and Varetto,model and the probabilistic NN without pattern normalization 1994;Lenard,Alam,and Madey,1995).This new methodol-are superior to the simple probabilistic NN model.The back-propagation NN model was not compared with the other three Address correspondence to Harlan D.Platt,Finance Group,College of Busi-in the final analysis based upon deflated data because it failed ness Administration,Northeastern University,413Hayden Hall,Boston,MA02115.to discriminate between bankrupt and nonbankrupt firms.Journal of Business Research44,67–74(1999)©1998Elsevier Science Inc.All rights reserved.ISSN0148-2963/99/$–see front matter 655Avenue of the Americas,New York,NY10010PII S0148-2963(97)00242-768J Busn Res Z.R.Yang1999:44:67–74Moreover,the linear discriminant analysis was better than the vector of class a,x a is the mean vector of patterns in class a,probabilistic NN without pattern normalization in correctly x is the mean vector of patterns in all classes,W is the variance classifying bankrupt firms.matrix,and B is the covariance matrix.Fisher (1936)discrimi-nant analysis searches for a linear subspace,of direction matrix M,such that the data,the X k ’s,project onto that subspace Comparison of Methodsgroup into well-separated clusters,one for each class.The Intergroup classification,including bankruptcy prediction,is distance between the projections of different classes are max-commonly undertaken with discriminant analysis that is ap-imized.The larger the between-class dispersion matrix (S B ),plied to a company’s accounting data.1Sample bifurcation the larger the dispersion of different populations and the is performed by using either linear or nonlinear projection smaller the total dispersion matrix (S T ),the more contractionoptimization methods or probability density estimation.These the densities of the populations.Let S *T and S *Bbe the total methods are described in the section below where the phrase and between-class dispersion matrices of the projections,3then a sample pattern denotes a pattern used in parameter fitting optimal M maximizes a cost function based on intraclass and during model construction,a validation pattern checks for interclass distances,as seen in the following equation for an over-fitting,and a test pattern verifies a model’s reliability.F-test (see Altman,1968).Projection optimization compares projection directions for various discriminant functions seeking an optimized projec-J (M)ϭ|S*B ||S*T |tion direction that minimizes misclassification among test pat-terns.Classification improves if the pattern swarm maximally Therefore,the larger the value of discriminant measurementseparates into distinct groups when they are projected in a J(M),the better the discriminant ability.particular projection direction (e.g.,the Euclidean coordinate Let A be S Ϫ1T S B and p be the rank 4of matrix S B (p рm Ϫ1).system).Fisher (1936)discriminant analyses are an example The matrix A has p nonzero eigenvalues,␮1Ͼ␮2Ͼ···Ͼof linear projection optimizations,whereas back-propagation ␮p Ͼ0,and p associated eigenvectors,w 1,...,w p .Then the NNs are nonlinear projection optimizations.matrix M whose columns are the w i ’s,maximizes J.As such,In contrast,the density estimate method compares the it is a solution to a discriminant analysis problem.With only contribution of a test pattern to the probability density func-one eigenvector,the method is simple discriminant analysis;tion of all the sample patterns and compares these contribu-with additional eigenvectors,the method is multiple discrimi-tions,which are multiplied by prior probabilities and loss nant analysis.The common form of a discriminant model is functions,in making a Bayesian decision.2Two factors influ-ence the contribution of a test pattern to the probability density z ϭc 0ϩc 1r 1ϩ···ϩc n r nfunction of a sample pattern.First is the relative spatial rela-tionship between the test pattern and a sample pattern.Second where,c i is a coefficient,r i is a variable (i.e.,usually a ratio)is the smoothing parameter ␴that is involved in density of the model and the dummy (bimodal)variable,z,is the estimation.The smoothing parameter weakens the contribu-model’s outcome.A vector containing all estimated coefficients tion of remote test patterns to the density function of a sample is called a particular projection direction.pattern.The probabilistic NN uses the density estimation The parallel distributed processing (PDP)group developed method.When the data space is highly overlapped between back-propagation NNs in 1986,(see Rumelhart and McClel-classes,projected patterns are harder to separate.Density esti-land,1986).It is a feedback net,which normally has three mation methods are more effective in these cases.layers:5an input layer,a hidden layer,and an output layer.Let S T and S B ,respectively,be the total and between-class The input nodes accept input signals,6the hidden nodes pre-dispersion matrices process the signals,and the output nodes process the prepro-cessed signals and output them.Connections between layers S T ϭ͚Ka ϭ1͚n aj ϭ1(u T x a j Ϫu T x a )2ϭu TWu are called weights.The basic principle of back-propagation NNs is continuous S B ϭ͚Ka ϭ1n a (u T x a Ϫu T x)2ϭu T Bu error feedback between the model and actual responses.Errorswhere,u T is the vector of desired coefficients,x a j is the pattern3S T and S B are calculated from the original data;S *T and S *B are calculated with projected data.4For a matrix,if there exists a maximum nonzero determinant,the number 1Discriminant analysis mentioned here does not refer to any particularmethod,such as Fisher discriminant analysis,Bayesian discriminant analysis,of rows or columns is equal to its rank.5Parameters of a model are updated in several non-linear equations.The or neural discriminant analysis.Unless otherwise specified,discriminant anal-ysis denotes the general method.final parameter solutions depends on many iterations of errors feeding back between outputs and inputs.2The method is similar to the k -nearest neighbors algorithm in which a particular test pattern is identified by taking the major class attribute of its 6For example,in failure prediction,a vector formed by several financial ratios.k nearest sample patterns.69 Neural Networks and Bankruptcy J Busn Res1999:44:67–74are calculated using the maximum likelihood rule.7The math-first layer,the input layer,accepts input patterns.The second ematical description of the negative log-likelihood is,layer contains several groups of pattern units,each of whichestimates the contribution of a particular pattern to the proba-LϭϪln⌫(c)ϭϪln⌸p(t n|c)ϭϪ͚ln p(t n|c)bility density function of each unit.The third layer is thewhere p(t n|c)is the conditional probability density of the n th summation layer in which the density estimate on each pattern vector t n under the condition c(or with class label,c).If t n is of each group is summarized.The fourth layer,the decision the response from a model corresponding to the error between layer,establishes a decision with Bayesian theory.the n th response and the n th target and c means one class,then The conditional probability function calculated in the sum-p(t n|c)indicates the possibility or mathematical probability of mation layer is described bythe error occurring at the n th company with class label c.Thelikelihood⌫(c)expresses the possibility of all the errors at p(x|c)ϭ͚M cmϭ1␻c m p(x|u c m)ϭ͚M cmϭ1␻c m΂␤c m2␲΃d/2·exp΂Ϫ␤c m|xϪu c m|22΃,the same time.The log function simplifies the calculationand is negative to meet the common optimization method:where M c is the number of the pattern units,␻c m is the mixing minimization.The objective is to train a net to minimize thecoefficient of the m th pattern units of the c th class,u c m is the negative log likelihood of the errors at which point the NNcenter of the m th pattern units of the c th class and␤c m is the has learned how to correctly classify.smoothing parameter corresponding to the m th pattern units Back-propagation NNs generally assume a normal underly-of the c th class.There are two assumptions with the traditional ing distribution of the errors,because the conditional proba-PNN(Specht,1990).One is equiprobable mixing of the pat-bility density is not easy to estimate when the distribution oftern units,the errors is unknown,␻c mϭ1M c .p(t n|c)ϭ1√2␲␴2exp(Ϫ|t nϪu c|22␴2)where u c is regarded as a target of class c.This assumption The other is the common smoothing parameterapplies to the output rather than to the input space(the␤c mϭ␤,1рmрM c,1рcрK,original data space).The error function of the back-propaga-tion NN then becomes a simple description,the least meanwhere K is the number of classes.The decision will be made square error,based on the Bayes ruleeϭ1p͚Ppϭ1|u pϪt p|2ϭ1p͚Ppϭ1|f(W1·g(W2·x p))Ϫt p|2d(x)ϭ␣,p␣l␣p(x|␣)ϭmax{p c l c p(x|c)},where,d(·)means a decision taken,p c is the prior probabilitywhere,P is the number of patterns;x p,the vector of p th inputof the c th class,l c is the loss function of the c th class. pattern;t p,the corresponding response vector;W1,the weightThe function of pattern units acts as a probabilistic density matrix connecting the hidden nodes with the input layer;W2,function.Here,the density estimate of a testing sample is the weight vector connecting the output nodes with the hiddencarried out on different classes.Identifying whether a testing layer;u p,the target of p th pattern;and f and g,non-linearsample belongs to the class␣or the class␥depends on the functions,such as a logistic function.The error is a function ofcontributions of the testing sample to the probability density weights.Training or learning updates the weights to minimizefunctions.Figure1gives an illustration.If a testing sample errors.falls in the range of᭝,this testing sample is most likely to Probabilistic NNs(see Specht,1990)employ Bayesian deci-belong to the highest probability density distribution␣rather sion-making theory based on an estimate of the probabilitythan␥because it has larger contribution to the probability density in data space.8A probabilistic NN has four layers.Thedensity function of␣than that of␥.The working principle of PNN can be seen from Figure2. 7If the total density function(probability function)is f(x;␪1,␪2,···,␪n),␪1,␪2,···,␪n are unknown and xϭ(x1,x2,···,x m)is a vector of sample values,In Figure2A,there are four pattern units centered at fourthen the likelihood function is L(·)ϭ͟Kiϭ1f(x i;␪i,␪2,···,␪n).If L(·)ϭmaxq1q2,···,q ndots—A,B,C,and D.The black dots belong to class␥andthe white dots belong to class␣.The radii of the gray areasL(X;␪1,␪2,...,␪n),then␪ˆ1,␪ˆ2,···,␪ˆn is a maximum likelihood estimate of␪1,␪2,···,␪n.centered at these four dots are the square root of the variance, 8The Bayesian decision-making theory is carried out by a comparison as which is used to enhance the probability contribution from followsthe samples of the same class and weaken the probability d(x)ϭ␣,if p(x|␣)p(␣)l␣Ͼp(x|␤)p(␤)l␤contribution from the samples of different classes.The square d(x)ϭ␤,if p(x|␣)p(␣)l␣Ͻp(x|␤)p(␤)l␤represents a testing sample,M.PNN calculates the probability where d(x)is the decision to be made,a and b are classes,p(a)and p(b)are contributions of this sample to the four pattern units and the prior probabilities,and l a and l b are loss functions connecting these twoclasses.sums the probability density functions(PDFs)according to70J Busn Res Z.R.Yang 1999:44:67–74Figure1.The working principle of PNN.different class labels of the pattern units.As shown in Figure Finally,PNN has been applied to many areas.However, 2B,the testing sample,M,is fed into the four pattern units—A,little attention has been paid to applying PNN to company B,C,and D,respectively.Each of them is an independent and failure prediction.There are two important factors determin-ing the performance of a PNN model:the positions of the identical Gaussian distribution,and each of them calculates acontribution from the testing sample,M,to its PDF.The filled pattern units and smoothing parameter.Pattern units are usu-rectangles to the left of the pattern units in Figure2B representally set by the training samples.The function of the smoothing these contributions to the PDFs.The larger the contribution,parameter has been discussed by Specht(1990),and it should the higher the rectangle.Then the PDFs are summed to differ-be optimally selected so as to minimize the misclassification ent class units␣and␥,respectively.The results from the rate.However,it is not so easy to select an optimal smoothing summation are called the conditional probabilities in statistics.parameter in practice.In order to obtain an approximate opti-Because the testing sample has more contribution to class␥mal value for the smoothing parameter,a complicated equa-than that to class␣,the filled rectangle aside the class unit␥tion has to be solved.Early work obtained an approximate is larger than that aside the class␣.So the decision is that optimal solution by hold-out technique(Musavi et al.,1996).With this hold-out technique,the training data set has to be the testing pattern M belongs to␥.It can be seen that PNNis simpler and more intuitive than BPNN and should not be split into two parts.All the samples in one of them are used regarded as a“black box.”as the centers of the pattern units of PNN,and the samplesFigure2.An illustration of the working principle of a simple PNN model.71Neural Networks and BankruptcyJ Busn Res 1999:44:67–74in another set are used for validation.The performance of a patterns have to be normalized before entering the model in PNN model is measured on the validation data set during the Specht’s (1990)original probabilistic neural network.The training.Musavi et al.(1996)varied the smoothing parameter normalization of a pattern is in the form of from 0.04to 0.3to search for a point with the minimum misclassification rate in the validation data set.x *ϭx x.If all the patterns,both the pattern units and the testing Data,Variables,and Deflationpatterns,have been normalized,the following equation can Platt,Platt,and Pedersen (1994)built an early warning bank-then be satisfied ruptcy model for the U.S.oil and gas industry.Their data for 122companies for the period 1984to 1989are used in this x i ·t c n Ϫ1ϭϪ|x i Ϫt c n |22.study.That study used a logistic model and a particular set of five ratios to classify oil and gas companies into bankrupt This leads to Specht’s original PNN.However,normalization and nonbankrupt groups.The five ratios are net cash flow to distorts the original character of a data space (Yang,1996).total assets,total debt to total assets,exploration expenses to For example,a linearly separable data space can become non-total reserves,current liabilities to total debt,and the trend linearly separable after normalization;as a result,the discrimi-in total reserves calculated asnant ability will be greatly reduced.Table 1presents the simulation results.With nondeflated change in total reserves (year 1to year 2)change in total reserves (year 2to year 3).variables,BP and PNN*are superior methods based on overall fit,whereas with deflated data,FISHER and PNN*are the The first four ratios are deflated to account for possible differ-top performers.As in Platt,Platt,and Pedersen (1994),deflat-ences over time in ratio values due to fluctuations in related ing ratios improves model classification for all methodologies.factors.The deflation process effectively removes factors that The poor showing of BP in finding bankrupt companies with are directly related to and thereby affect the value of the target deflated data is surprising and questions the method’s use-variables.Dividing oil prices or interest rates,as appropriate,fulness.Specifically,BP failed to model the deflated data set;into the relevant variables formed deflated ratios.Conse-that is,it failed to distinguish the bankrupt companies from quently,there are two separate data sets:one with deflated nonbankrupt companies.Given the BP model’s failure to dis-ratios and one without deflation.This study of alternate meth-criminate between bankrupt and nonbankrupt firms,it is fair odologies accepts the Platt,Platt,and Pedersen (1994)model to exclude this model from further consideration.specification as given and does not seek a best set of financial Focusing attention on model results based on the deflated ratios for each methodology.This approach permits the meth-data,three ␹2tests of independence were conducted to assess ods to be compared.It is not designed to yield the best early possible differences among models in classification accuracy.warning model that one would use in actual applications.First,a ␹2test of independence was conducted that compared The 122patterns (i.e.,companies)are randomly divided the overall results from the three models.This test yielded a into three sets:the training data set,the validation data set,␹2ϭ4.86with 2df and a p -value beyond the .10level.These and the testing data set.The training data set contains sample results suggest that there are marginally significant differences patterns,the validation data set contains validation patterns,among the three models.Supporting this finding,the ␹2test and the testing data set contains testing patterns.In the train-of independence based upon the three models’results classify-ing data set,there are 33nonbankrupt companies and 11ing the bankrupt firms found significant differences among bankrupt companies.In the validation data set,there are 26the three models,with ␹2ϭ6.51with 2df and a p -value nonbankrupt companies and 14bankrupt companies.In the beyond the .05level of significance.However,the ␹2test testing data set,there are 30nonbankrupt companies and of independence based on correctly classifying nonbankrupt eight bankrupt companies.The training data set is used for firms was not statistically significant.fitting parameters of a model during model construction,the Given that the models differ marginally with respect to validation data set is used to check for over-fitting (normally used in the back-propagation NNs),and the testing data set both overall classification accuracy as well as significantly in determines the reliability of a model.classifying the bankrupt firms,statistical tests comparing pairs of models’classification accuracy were conducted for the over-all results as well as for the bankrupt firms.McNemar (1947)Simulation Results␹2tests for paired sample nominal scale data were conducted when n,the number of companies in the analysis,was greater Four different methods are tested:FISHER denotes Fisher than ten.Otherwise,binomial tests are required (see Zar,discriminant analysis,BP represents back-propagation NN,1974).These tests analyze a 2ϫ2contingency table where PNN means probabilistic NN,and PNN*is probabilistic NN without patterns normalized.Both the training and testingthe joint classification outcomes of two models are tallied.72J Busn Res Z.R.Yang 1999:44:67–74Table1.Number and Percentage of Companies Correctly ClassifiedNondeflated Data Deflated DataOverall Nonbankrupt Bankrupt Overall Nonbankrupt Bankrupt(nϭ38)(nϭ30)(nϭ8)(nϭ38)(nϭ30)(nϭ8)FISHER2720733267 71%67%88%87%87%88% BP2824430300 74%80%50%79%100%0% PNN2524126242 66%80%13%68%80%25% PNN*2824432275 74%80%50%84%90%63%The null hypothesis tested is that there is no difference in finally PNN.In this ranking,FISHER was placed over PNN*due to the better classification accuracy in the paired sample classification accuracy between the two models.None of the␹2tests for the total sample reached statistical comparison between these two models.Thus,the Fisher dis-criminate analysis is best at minimizing Type2error—that significance.Table2reports the value of binomial tests com-paring pairs of models using a2ϫ2table of joint model is,the misclassification of bankrupt companies.This criticalresult supports Altman,Marco,and Varetto(1994). classification outcomes.Because a binomial test is based ona random variable that has only two possible outcomes,the All discriminant results in Table1are lower than expected(about90%)based on results reported by Platt and Platt outcome designated as a“success”is arbitrary.For this analy-sis,a success was said to occur when one(not both)of the(1990).Yang and James(1997)investigated this issue.Theyfound that if the distributions of two groups of companies models misclassified a bankrupt company.The formula forthe probability of exactly X successes out of n trials is:are highly overlapped in a data space,that probabilistic NNand Fisher discriminant analysis have similar discriminantresults with extreme misclassification rates of about50%.An P΂X n΃ϭn!x!(nϪx)!p x(1Ϫp)nϪx overlap test statistic developed by Yang and James(1997)iswhere X is the number of successes,n is the number of trials,Xϭ1N͚Nnϭ1(1Ϫ|p(x n|␣)Ϫp(x n|␤)|max{p(x n|␣),p(x n|␤)})and p is the probability of success.For the tests conductedbelow,the overall probability of success across the three mod-els,pϭ.58,was used.The number of trials for all tests was The critical value of X will approach zero for a nonoverlapping 8,the number of bankrupt companies.The null hypothesis data space and approach one for a highly overlapped data is rejected if P(X/n)Ͻ.05.space.Table3presents the overlap test results.The overlap Table2reveals that the classification accuracy of the Fisher degree is much higher with nondeflated ratios.That is,defla-discriminant analysis model dominates PNN(with patterns tion improves the discriminant ability of a bankruptcy predic-normalized).The FISHER model,however,does no better tion model because the overlap within different distributions overall than PNN*(without patterns normalized).Moreover,has been weakened.Although deflation does not greatly im-the PNN*model is not statistically significantly better than prove the overall results(i.e.,from47%to45%),the improve-the PNN model in correctly classifying sample firms.Thus,ment in the bankrupt companies group was very significant it appears that the relative ranking of the three models in(from88%to68%).terms of classification accuracy would be FISHER,PNN*,andSummaryTable2.Binomial Tests for Model Comparisons Using Deflated DataThis study has briefly reviewed two types of discriminantBankrupt Group models,projection optimization and probability density esti-Method(8companies)FISHER-PNN[6]Table3.Overlap Test(a value of unity is extreme data overlap)(0.01)FISHER-PNN*[3]Bankrupt Nonbankrupt(0.25)Companies Companies Total PNN*-PNN[5](0.05)Nondeflated data88%31%47%Deflated data set68%36%45% The number of successes in brackets;p-value in parentheses.73Neural Networks and BankruptcyJ Busn Res 1999:44:67–74Korobow,L.,Stuhr,D.P.,and Martin,D.:A Probabilistic Approach mation.Four different methods,Fisher discriminant analysis,to Early Warning of Change in Bank Financial Conditions.Federal the back-propagation NN,the probabilistic NN,and the prob-Reserve Bank of New York Monthly Review 58(1976):187–194.abilistic NN without the patterns normalized,have been ap-Kryzanowski,Lawrence,Galler,Michael,and Wright,David W.:plied to bankruptcy prediction using financial ratios (nonde-Using Artificial Neural Networks to Pick Stocks.Financial Analyst flated and deflated)from the U.S.oil industry.The study Journal 49(1993):21–27.examines the differences among these four methods.Experi-Kuan,Chung-Ming,and Liu,Tung:Forecasting Exchange Rates Us-ments show that the probabilistic NN without patterns nor-ing Feed-Forward and Recurrent Neural Networks.Journal of Applied Econometrics 10(1995):347–364.malized and the back-propagation NN models obtained the highest classification accuracy overall by using nondeflated Lenard,Mary Jane,Alam,Pervaiz,and Madey,Gregory R.:The Application of Neural Networks and a Qualitative Response Model data,primarily because they accurately predicted nonbank-to the Auditor’s Going Concern Uncertainty Decision.Decision rupt firms.However,in bankruptcy prediction research,the Sciences 26(1995):209–227.Type II error (misclassifying a bankrupt firm)is more costly;McNemar,Q.:Note on the Sampling Error of the Difference between thus,more attention should be paid to this type of error.It Correlated Proportions or Percentages.Psychometrica 12(1947):is important to note that the Fisher discriminant analysis 153–157.produced the best classification results with deflated data.In Musavi,M.T.,Qiao,M.,Davisson,M.T.and Akeson,E.C.:Classifica-fact,regardless of data format,the Fisher discriminant analysis tion of Mouse Chromosomes Using Artificial Neural Networks.more accurately predicted the status of bankrupt companies The 1966IEEE International Conference on Neural Networks 2(1996):852–857.(see Appendix A).Pantalone,Coleen,and Platt,Marjorie B.:Predicting Commercial Another important finding is that deflation improves the Bank Failure since Deregulation.New England Economic Review discriminant ability of bankruptcy prediction models,which (July/August 1987):37–47.supports Platt,Platt,and Pedersen’s (1994)earlier work.These Platt,Harlan D.,and Platt,Marjorie B.:Development of a Class of results have been verified based upon ␹2statistical tests as Stable Predictive Variables:The Case of Bankruptcy Prediction.well as with the overlap test.Journal of Business Finance &Accounting 17(Spring 1990):31–51.Platt,Harlan D.,Platt,Marjorie B.,and Pedersen,Jon:Bankruptcy ReferencesDiscrimination with Real Variables,Journal of Business Finance &Accounting 21(June 1994):491–509.Altman,Edward I.:Financial Ratios,Discriminant Analysis and the Prediction of Corporate Bankruptcy.Journal of Finance 23(Sep-Rumelhart,D.,and McClelland,J.:Parallel Distributed Processing:tember 1968):589–609.Exploration in the Cognition ,MIT Press,Cambridge,MA.1986.Altman,Edward I.,Haldeman,R.,and Narayanan,P.:ZETA Analysis:Salchenberger,Linda M.,Cinar,E.Mine,and Lash,Nicholas A.:A New Model to Identify Bankruptcy Risk of Corporations.Journal Neural Networks:A New Tool for Predicting Thrift Failures.of Banking and Finance 1(1977):29–54.Decision Sciences 23(1992):899–916.Altman,Edward I.,Giancarlo,Marco,and Varetto,Franco.:Corpo-Specht,Donald:Probabilistic Neural Networks.Neural Networks 3rate Distress Diagnosis:Comparisons Using Linear Discriminant (1990):109–118.Analysis and Neural Networks (the Italian Experience).Journal Worzala,Elaine,Lenk,Margarita,and Silva,Ana:An Exploration of of Banking and Finance 18(1994):505–529.Neural Networks and Its Application to Real Estate Valuation.Journal of Real Estate Research 10(1995):185–201.Barth,J.R.,Brumbaugh,R.D.,Sauerhaft,D.,and Wang,G.:Thrift-Institution Failures:Causes and Policy Issues.Research Working Yang,Z.R.:Does the Probabilistic Neural Network Need Normaliza-Paper No.117,Office of Policy and Economic Research,Federal tion.Working Paper No.10,ISSN 1366-3178,1996.Home Loan Bank Board,1985.Yang,Z.R.,James,H.:The Overlap Test and Its Applications.The Coats,P.,and Fant,L.:Recognizing Financial Distress Patterns Using Second International Symposium on Intelligent Data Analysis (AAAI,a Neural Network Tool.Financial Management 22(1993):142–ACM SIGART,BCS SGES,IEEE SMC and SSAISB),Birkbeck 155.College,London.1997.Fisher,R.A.:The Use of Multiple Measurements in Taxonomic Zar,Jerrold H.:Biostatistical Analysis ,Prentice-Hall,Inc,Englewood Problems.Annual Eugenics 7(1936):178–188.Cliffs,NJ.1974.Kaastra,Iebeling,and Boyd,Milton S.:Forecasting Futures Trading Zmijewski,M.:Methodological Issues Related to the Estimation of Volumes Using Neural Networks.Journal of Futures Markets 15Financial Distress Prediction Models.Journal of Accounting Re-search 22(1984):59–82.(1995):953–970.。