Known-item retrieval on broadcast TV
J.A. List, A.R. van Ballegooij, A.P. de Vries Information Systems (INS)
INS-R0104 April 30, 2001
Report INS-R0104
ISSN 1386-3681
CWI
P.O. Box 94079
1090 GB Amsterdam
The Netherlands
CWI is the National Research Institute for Mathematics and Computer Science. CWI is part of the Stichting Mathematisch Centrum (SMC), the Dutch foundation for promotion of mathematics and computer science
and their applications.
SMC is sponsored by the Netherlands Organization for Scientific Research (NWO). CWI is a member of ERCIM, the European Research Consortium for Informatics and Mathematics.Copyright ? Stichting Mathematisch Centrum P.O. Box 94079, 1090 GB Amsterdam (NL) Kruislaan 413, 1098 SJ Amsterdam (NL)
Telephone +31 20 592 9333
Telefax +31 20 592 4199
Known-Item Retrieval on Broadcast TV
Johan List,Alex van Ballegooij,Arjen de Vries
{j.a.list,alex.van.ballegooij,arjen.de.vries}@cwi.nl
CWI
P.O.Box94079,1090GB Amsterdam,The Netherlands
ABSTRACT
Many content-based,multimedia retrieval systems are based on a feature-oriented approach to querying,
mostly exposing a?xed set of features(introduced at design time)for querying purposes.This restriction
to a limited set of features is problematic for two reasons:it restricts the expressiveness at the semantic
level,and it seems unfeasible to obtain(a-priori)a su?ciently powerful set of features for all possible
queries.
We describe an alternative approach where users specify precisely the distinguishable characteristics of the desired result set.In this query process,the user?rst describes a representation of the content(based
on a feature or collection of features)and then tells the system how to apply the representation in the
search.
Our prototype video retrieval system allows the expression of such queries as a sequence of operations, on MPEG video and audio streams,that can be executed on our database system.While the low-level
decompression stage is implemented in an imperative programming language,the actual retrieval approach
is expressed in declarative database queries.We assessed this system with a case study in known-item
retrieval on broadcast video streams:detecting news bulletins in the stream,with the help of both audio
and video information.
2000Mathematics Subject Classi?cation:68P20,68U99
1998ACM Computing Classi?cation System:Information Search and Retrieval(H.3.3),Multimedia Infor-
mation Systems(H.5.1)
Keywords and Phrases:multimedia retrieval systems,multimedia database systems,known-item retrieval
Note:Work carried out under INS1MIA project and the DRUID project and the report has been submitted
as article to CBMI2001
1.I NTRODUCTION
The most challenging area of multimedia retrieval research is closing the semantic gap,https://www.doczj.com/doc/0613524548.html,bining and mapping low level media-speci?c features in an intelligent way to high-level concepts.Many multimedia retrieval systems focus on a feature-oriented approach to querying.A single representation of the content of multimedia objects is decided upon a-priori.Because an optimal representation for all queries does not seem to exist,researchers attempted to use a series of representations(termed a‘society of models’in[1]), and select a good representation based on user feedback,see e.g.[1,2].
Choosing the right representation automatically is very important for naive users with generic queries. But,guessing the right query representation from user feedback is a very dif?cult problem,and has not been solved to a satisfactory level.The question arises whether the user can play a larger role in the‘articulation’of the query than simply saying yes or no to retrieved objects.
This paper describes an approach in which users de?ne their information needs precisely in terms of the distinguishing characteristics of desired responses.The query does not necessarily depend on a single rep-resentation of content:it can be constructed from several representations,possibly containing information from different media types.The analogy in image retrieval is our work on the‘Image-Spotter’[3],allowing its users to designate the regions of interest in an image.
Consider for example a user searching for video segments with a space shuttle in orbit over the Earth. Looking at the characteristics of such segments,the user may come up with the following distinguishing characteristics:
2.Multimedia Retrieval Systems2
Figure1:Example Space Shuttle Search Image
–the narration track(if there is any present)is likely to contain words such as‘space shuttle’,‘orbit’,‘earth’;
–there will probably be a large amount of dark colors present in such a segment;
–the space shuttle is mostly white;
–Earth itself is characterized as a collection of blue,green,brown and white colors;
–there is a certain spatial relation between the space shuttle and the Earth.
The user speci?es the representations for(several of)the characteristics mentioned above,and instructs the system on how to use these representations to answer the query.Some of these characteristics may be complex and dif?cult to specify by hand,such as the shape of the space shuttle or the variety of colors present in the Earth.Then,an example image such as the one shown in Figure1serves as a reference for the system,to determine these characteristics automatically.
We present a work in progress to support the query process sketched in the(hypothetical)example above. We designed and implemented a prototype video retrieval system,and tested its merits with a case study in known-item retrieval.This case study comprised?nding the delimiting segments of Dutch news bulletins, from broadcast video streams containing commercials before the start and after the ending of the news bulletins.
The structure of the rest of the paper is as follows.Section2discusses relevant work in the?eld of multimedia retrieval systems,followed by Section3containing our approach.Sections4and5describe our prototype system architecture and the case study respectively.Sections6and7present our experimental results,conclusions and directions for future research.
2.M ULTIMEDIA R ETRIEVAL S YSTEMS
For the answering of queries,many multimedia retrieval systems focus on content-based retrieval:the analysis and extraction of low-level,media-speci?c features,followed by a similarity search in the feature spaces available.We have looked at two speci?c cases of multimedia retrieval systems:video and audio retrieval systems.
The?rst video retrieval systems considered video retrieval as a special case of image retrieval.An example of such a system is QBIC[4].In QBIC,video data is?rst segmented at the shot level,after which key frames are chosen or generated(mosaic frames in the case of shots with panning motion).Object detection was achieved through analyzing the layered representation of video:the detection of regions with coherent motion.
VideoQ[5]introduced the concept of video objects.Video objects are sets of regions which display some amount of consistency,under certain criteria,across several frames.As in QBIC,video data is?rst segmented at the shot level after which the global background motion of each shot is determined.Color-, edge-,and motion-information of regions within the shot is then extracted to track possible video objects across several frames.
Query interfaces for multimedia retrieval can be distinguished into textual or visual query interfaces. Visual query interfaces mostly comprise query by feature or feature combinations(local or global features) and query by sketch or example.The VideoQ system allows the sketch of a motion trajectory for a query object as well,to capture the motion aspect present in video data,and a time duration for the searched video segment(either an absolute duration or an intuitive measure such as’long’,’medium’or’short’).
3.Our Approach3
Figure2:System Architecture
An interesting approach to querying video material,which comes closest to what we want to achieve, is described in[6]and is built on top of the VideoQ system.Querying is performed with semantic visual templates,i.e.a set of icons or example objects that form a representation of the semantic the user is searching for.An icon is an animated sketch,composed of graphical objects which resemble the real-world objects.For searching the system,features are extracted from the icons or example objects and a similarity search is performed.
A similar content-based retrieval approach has been followed in audio retrieval research[7].Muscle-?sh[8]extracts features such as loudness,pitch and brightness.In querying the audio database,the user can specify an example sound from which features are extracted and used during query processing. SpeechSkimmer[9]is a system developed for interactively browsing or skimming of speech.The system uses time compression processing to enable users to listen to speech several times the speed of real-time, based on the notion that human speech can be understood much quicker than the normal speech rate. Speech sounds can be segmented by using information about the speaker’s pitch or the use of speaker identi?cation.
An important feature of audio retrieval systems is the query by example feature,where an example sound segment is provided to the system for querying.Instead of using example sound segments,the system developed at the University of Waikato[10]uses acoustic input for querying:the user can hum a tune.Audio?les present in the system are then analyzed for tunes similar to the hummed tune,using pitch extraction or string-matching algorithms(in the case of the Waikato system).
3.O UR A PPROACH
A problem is that in many content-based retrieval systems,a standard set of features is present(introduced at design time)for the similarity searches.We deem it impossible to determine a-priori a standard set of features which can be used for answering all queries possible.Moreover,there are only so many objects, concepts or relations to describe as the used feature space(combinations)allow to express.Most impor-tantly,it is impossible to a-priori determine which feature space best captures higher-level concepts for all possible domains,a standard feature-set may not be suitable to exploit available domain knowledge. Since domain speci?c search strategies generally result in better answers than generic approaches,this is an important issue.
Based on the above,we also conjure that,in the case of a layered or integrated approach to multimedia retrieval(examples include[11,12,13]),the restrictiveness in feature spaces present causes restrictiveness at the higher levels.
In our eyes,users can contribute to crossing the semantic gap with a top-down oriented approach,by taking advantage of the domain knowledge of the(expert)users themselves.A multimedia retrieval system should therefore offer the users the possibility to de?ne their information need in terms of distinguishable characteristics of desired responses.
Note that we do not completely dismiss content-based retrieval based on media features.At some point during the query process,media features and the accompanying similarity search will be necessary.We only question the manner in which features are used for retrieval in many systems today.
To test and illustrate our approach,we built a prototype video retrieval system,of which the architecture is shown in Figure2.The lowest level of the system consists of the raw bit streams(such as video streams,
4.System Architecture4 audio streams and other media)that are stored in a multimedia database system.
The information that is effectively stored in the database system is the information needed to both de-compress and decode the audio and video data of the MPEG bit stream.The next abstraction level consists of media-speci?c analysis or feature extraction elements(such as frames from a video stream or sample data from an audio stream).We propose that the construction of the elements,in this second level of abstraction,the frames and sample data in itself can be regarded as a view on the low-level bit stream. The video frames and sample data form the basis for feature extraction and analysis,which in our system is written in a higher-level database interface language and consist of sequences of operations on database tables.This is what allows the exact feature extraction algorithms to be chosen at query-time,thus allowing for the most appropriate ones with respect to the domain to be used.
The system as described here is not a complete video retrieval system.It should be considered as a framework consisting of basic operations that requires an actual retrieval model to be built by means of a query-de?nition for each separate retrieval problem.The abstract interface that is a result of the use of a database system allows such systems to be build relatively easily and quickly.Nevertheless it does place a larger burden on the users.In order to reduce this burden we plan to construct an intermediate layer between the system and the user.This layer will act like a domain-independent thesaurus.This thesaurus component can offer high level primitives(such as’list all speech segments in a certain stream’together with a description of the appropriate media-representations needed for answering this sub-query).
The prototype system as it stands now is aimed at expert users with knowledge of both video-and audio processing.Our intent is to extend the system further for making it usable by ordinary users(see Section7).
4.S YSTEM A RCHITECTURE
For storage of the information needed for both decompression and decoding of the video-and audio streams (see Figure2),we used the Monet database system[14].The main advantage of using database systems for information retrieval task is that they“offer?exible declarative speci?cation means and provide ef?ciency in execution,through algebraic optimization in the mapping process from speci?cation to query execution plan”[15].
4.1The Monet Database System
The Monet database system is a main-memory,parallel database system developed at CWI[14].Its data model is based on Binary Association Tables(BATs),which are two-column tables consisting of a head and tail value.Monet offers a?exible algebra to manipulate the BATs present.
The design goals of Monet included high performance and extensibility.High performance is gained through the execution of operations in main memory,the exploitation of inter-operation parallelism,the frequent use of bulk operations which optimizes cache usage and a simple data model.Extensibility is offered through modules,in which application programmers can de?ne new data types(called atoms), commands,operators,and procedures.Procedures are written in MIL(Monet Interface Language)which is a higher-level language de?ned for user-level interaction with the database.
At this point,MIL plays the role of a higher-level speci?cation language.We plan to introduce a layer on top of MIL with a query language more suited for the multimodal access and https://www.doczj.com/doc/0613524548.html, will then be used as an intermediate layer for communication between user interface components and the retrieval system as a whole.
4.2Multimedia Extensions
For our experiments an MPEG video and an MPEG audio module were implemented in C.Decoding of the video and audio streams were implemented as commands whereas much of the feature extraction and analysis operations were written in MIL.
The video module consists of a set of database commands that allow both the decompression and if needed decoding of a given range of frames from an MPEG?le.This is needed since we want users to have access to all data in the video stream.We consider storing complete decompressed video streams unfeasible because of the enormous amount of data present in such a stream.A rough calculation indicated that10 hours of video(approximately5GB of MPEG I data)requires500GB for storage of the decompressed video,audio and basic indexing information.
The database commands allow users to access both compressed-domain1and image-domain data to search in the video stream.
Analogous to the video module,the audio module primarily consists of a set of database commands that allow both the decompression and decoding of a given range of audio frames from the audio stream in an MPEG?le.Even though the size of audio-data is only a fraction of the video data present in a stream many of the same database-size restrictions hold and the storage of all decompressed data is unfeasible.In addition to these decompression and decoding functions the audio module contains a set of standard signal processing functions that users can use to extract interesting features from the audio data.
The nature of the system encourages(expert)users to de?ne custom algorithms,but we assume certain operations to be of such importance that optimized implementations are provided in the form of database operations.Examples of such optimized operations are color space conversion and histogram construction, which are relatively low-level and computationally expensive operations.
5.C ASE S TUDY
In the case study used to illustrate our approach,we focused on a speci?c known-item retrieval problem in broadcast news,namely to?nd the starting and ending time indices of news broadcasts.This speci?c case study was chosen to complement an automated news recording system that is in use for the construction of a data warehouse of news broadcasts[16].Until now,the determination of the exact beginning and end of each recorded news broadcast had to be done by hand.
5.1Characteristics of News Broadcasts
News broadcasts are relatively structured and are often characterized by the following properties:
Speci?c”delimiter”parts Broadcast news programs often start with a speci?c frame or set of frames with a constant layout,see Figure3.In our case the?rst frame of the opening sequence shows the NOS2logo on a black background.The last frame of the end sequence shows a globe surrounded by some graphics and a caption that announces the NOS website[17].A similar argument can be made for the audio part of the broadcast.Recognizable tunes are played during the start and end sequences.Additionally,the start and end sequences are accompanied by short periods of silence. Speech-richness of broadcast news News broadcasts often contains mostly speech,whereas the surround-ing commercials are characterized by a more rapid mix of both speech and music.Also,commercial segments appear louder to the human ear in comparison to speech signals.
High cut frequency of commercial material Commercial material is often allocated a certain slot of time during which a certain amount of commercials is shown.The average duration of a commercial is 30seconds in which all product-related information must be placed.Hence,we can expect a higher level of action(or cut frequency)during commercials.
Note that these characteristics can be collected by watching a collection of news broadcasts.
5.2Audio and Video Feature Extraction
Given the characteristics of our domain as described in Section5.1,we chose to use a number of suitable feature extraction and matching algorithms.Note that we chose these speci?c features because we found them to be applicable to this speci?c problem,the system in no way prescribes or limits us to the use of the features presented here.In choosing suitable feature extraction and matching algorithms,we focused on the the speci?c delimiter parts of Dutch news broadcast.Also note that two media types(video and audio) are used in analyzing the streams.The main idea is that,given the problem,simple,media speci?c features are suf?cient for solving the problem when combined during retrieval.
1What is usually referred to as’compressed domain’data is in reality already decompressed.MPEG video data is conceptually decoded in two stages.Firstly Huffman decompression of the bit stream results in motion vectors,and DCT domain pixel descriptions, this is the data usually referred to as compressed-domain.Secondly this data can be used to construct the actual images contained in the video stream.
2NOS stands for”Nederlandse Omroep Stichting”,the Dutch broadcast organization.
Figure 3:The begin ?and end frames of a news broadcast ?Note that the box depicted in the begin image indicates the bounding box used in the image matching step and is not
part of the actual image.
The ?rst step of the algorithm is the analysis of the audio stream,and speci?cally the occurrence of high energy parts in the signal.The delimiter parts of news broadcasts are accompanied by speci?c tunes that are played.Both these tunes show up in the analysis as periods of high energy.
The time index generated during analysis of the audio stream is used by the video module to prune the video stream search space.The frame segments surrounding the high energy periods are decompressed and decoded and analyzed for the occurrence of the speci?c delimiter frames.
Audio Features Feature extraction of the audio signal uses compressed MPEG audio data [18].Using compressed MPEG audio data for audio analysis has proven to be an ef?cient method giving good results compared with analysis on time-domain waveform data [19,20].MPEG audio compression is based on perceptual audio coding,which uses the characteristics of the human auditory system in order to gain higher compression ratios.
In order to analyze the audio signal,the approach in [20]was followed where a mean sub band vector is calculated for each frame.A frame is a collection of sub band vectors,each specifying the spectral content of 32raw input samples.In Layer II MPEG audio a frame consists of 3sub frames,each containing 12groups of 32-element sub band sample vectors.A mean sub band vector can then be calculated as follows:
M [i ]= i =1S t [i ]3?12
,i =1..32The collection of mean sub band vectors is then further analyzed on energy content,which indicates the loudness of the frame,and is used in detecting periods with high energy in the audio signal.Note that the concept of high energy can indicate a variety of possibilities.A high energy segment can include music,loud speech,or loud background sounds.So care must be taken not to assume too much on the basis
var beginImages := shots.getKeyFrames_First(fname).rankImages("beginImage_2001");var endImages := shots.getKeyFrames_Last(fname).
rankImages("endImage_2001");
var intervals := high_energy_thresh(high_energy(rmsValues), int(10), dbl(0.85));
var rmsValues := [rms](mean_vectors(decodedStream);
var decodedStream := decode_audio_stream(fname);
var shots := bat(int, int);
# fname contains filename of MPEG stream to be examined
intervals@batloop { shots.insert(getShotsForTimeSlice(fname, $h, $t); }var begin := beginImages.reverse.fetch(0);
var end := endImages.reverse.fetch(0);
printf("The news starts at frame %d and ends at frame %d\n", begin, end);Figure 4:
(a)Query-Graph for start and end detection.(b)A snippet of the corresponding MIL query.
of high energy information,als analysis windows must be chosen with care to ensure the best possible classi?cation.Analogous to the high energy variant,periods with low energy in the audio stream can be detected.Low energy periods can indicate silence segments.
Video Features The video data consists of MPEG-I video streams[21].The smallest unit in a video stream is a frame,in other words a video stream is a sequence of images.In order to provide some structure in a large video stream it is useful to segment the stream into meaningful sections.There are two levels of video segmentation,the?rst level is the shot level,this essentially constitutes a piece of video shot by a camera in one consecutive go.A higher level of grouping could be managed in the form of scenes,scenes are groups of shots constituting a semantic unit,for example a distinct part of a story.
A simple algorithm for?nding the speci?c delimiter parts(see Subsection5.1)in the video stream is as follows.First the video stream is segmented at shot level.The collection of detected shots is then analyzed further to give us key frames.The key frames are then given to an image comparison operation,where each key frame is compared with a given image to?nd the one that matches best.
Indexing the videos at shot level is good enough for rough video segmentation.Shot level segmentation can be achieved in a number of ways,amongst which a pair-wise image comparison of consecutive frames based on color histograms[22].Given the fact that in our problem domain,we are only looking for the hard cuts(the delimiter frames are suf?ciently different from surrounding frames)the computationally rel-atively inexpensive approach of comparing image color histograms suf?ces.Histogram based cut-detection performs equally well to other more advanced cut-detection algorithms when hard-cuts are the subject of detection[23].Although certain color spaces perform better than others empirical examination of the prob-lem domain has shown that using only the luminance component of the images suf?ces to detect the fast majority of cuts in the data[24].More importantly,this relatively simple approach has shown to have near-zero miss rate in our data-set.Of course,there is a trade-off against a higher false-hit rate.
The second step comprises key frame selection.Once again,for this problem the simplest approach is best,since we are either looking for the beginning of the news broadcast(which always starts at the beginning of a shot)or the end of the news broadcast(which ends at the end of a shot).Simply selecting either the?rst or last frame of a shot as the key frame is exactly what we want.
The third step is image comparison.Since we exactly know what we are looking for and the start and end sequence is always the same computer generated animation,a basic pixel-wise image comparison function is adequate.The exact algorithm measures the difference between corresponding pixels on a macro block level.For the ranking of the possible start frames only the macro blocks contained in the bounding box depicted in Figure3are used essentially realizing a spatial constraint.For the ranking of the possible end frames the complete image is used.The difference between macro blocks is calculated as follows:
MB di?(a,b)= 256i=1min((a i?b i)2255,100)
The differences are combined in a vector,one dimension per macro-block.The length of this vector is used to express the difference between two images.
6.R ESULTS
The test-set used for the experiments was composed of several news broadcasts recorded at different times during the day over a period of a few days,stored in MPEG-1format.The video content was encoded in standard PAL format,25frames of352x288pixels per second.The audio content was encoded with MPEG Layer II compression at44.1KHz16bit sampling.
The?rst algorithm we used was segmentation of the video data at shot level,followed by selection of key frames and a full-frame comparison of the key frames and example images.Although this approach has high precision,it is also very computationally expensive.
We attempted to realize a speed-up in searching the video stream by using the low energy information from the audio stream.Assuming that we would be able to?nd the silences that are present at the beginning and end of respectively the start and end sequence of the news broadcast.The introduction of the energy content analysis of the audio stream cut the execution time with roughly a factor of2.Although we were able to identify the silences we are interested in,we noticed that a large number of short silences is present in the test set.Thus,the percentage of video material that has to be examined even after only
Method Start Frame End Frame Execution Time Video full-frames80%100%2?D
Video full-frames and low-energy Audio selection80%100%(40%?2?D)+(1
10D)
Video full-frames and high-energy Audio selection80%100%(5%?2?D)+(1
10D)
Video spatial and high-energy Audio selection100%100%(5%?2?D)+(1
10
D)
Figure5:Results of the experiments
selecting low energy periods is relatively large,about40%.Limiting the number of low energy periods,
by changing the selection of the duration of those low energy periods,resulted in unsuccessful detections. After examination of the test set material,we discovered that the silence periods accompanying the start of the news broadcasts were not of a constant duration.
Selecting high energy periods proved to be a better approach.We used the high energy information to mark possible musical segments such as those played during the start and end of the news broadcast.We discovered that this feature prunes the video search space signi?cantly better than the low energy variant. Another advantage is that the duration of the start and end musical segments are constant.This allowed us
to reduce the collection of detected high energy periods with a selection based on duration,reducing the amount of video data that needs to be examined to approximately5%of the total duration.As a result we realized a speed-up of roughly a factor10.
Using the high energy information,all end frames were identi?ed correctly,but we still experienced problem with identifying the start frames in a few cases.We discovered that the misses were a result of the used image ranking algorithm.To remedy this,we added a spatial constraint to the image compari-son for the start frame,by limiting the comparison to the bounding box of the logo.This counteracted the domination of the background color on the image comparison results,solving the mis-identi?cation problem.
The fact that we obtained a100%success rate over our test set is easily explained:
–the test set was relatively small,consisting of10items;
–this retrieval problem is a relatively simple problem because the characteristics of the data set are clearly de?ned;
–the computer generated animations from the start and end sequences are easily distinguished from other natural images present in such a video stream.
Another important characteristic of the data present in the test set is that the image quality is rather high. When lower quality recordings are to be tested it is likely that the algorithms will perform less well.
A more generic measure for gained performance by using the combined audio-and video information could be expressed as follows.Let D be the duration of the video stream.We regard D to be the same for the audio stream since we are regarding entire video streams.The decompression,decoding and analysis of the entire video stream has a duration of2?D.The low and high energy analysis of the audio-stream can
be performed in approximately1
10?D seconds.Therefore if we know the fraction S of video information
that needs to be examined after the audio analysis,we can approximate the total execution time as(S?2?
D)+(1
10D).Table5shows the analysis information and execution times from our different attempts.
To improve on the performance of the retrieval algorithm,other feature extraction algorithms could be introduced.Speci?cally introducing extraction algorithms for the other characteristics mentioned in Section5.1may prove to be useful.A possibility is to implement a music/speech classi?er(as in[25]), which provides a more robust classi?cation of music and speech segments and hence a more robust marking of possible start and end points of the news broadcast.Additionally a speech/music classi?cation,combined with the fact that large parts of news broadcasts are often characterized by speech,may contribute to marking possible starting and ending points in the stream more precisely.
A second category of improvements is the optimization of the system itself.The code base used for this system is a naive implementation of MPEG decoding.We have observed optimized MPEG-decoder programs to decode the same streams at least10times faster.Therefore,ef?ciency gains can be expected in that area as well.The possibilities the Monet database system offers for parallelism can provide signi?cant speed-up as well.
7.Conclusions and Future Research9 7.C ONCLUSIONS AND F UTURE R ESEARCH
Our experiments show that the approach we discuss is viable although our implementation still lacks per-formance.
The developed prototype system allowed the articulation of the query graph presented in Figure4in higher-level scripts and in our eyes proved to be an ef?cient method for expressing video retrieval queries. Admittedly,the solution for this speci?c problem is not generic;however,the framework we propose is generic.We feel that this shows that the ability to make use of domain knowledge allows the usage of simple algorithms,while obtaining a high precision.However decent user support in query formulation will have to be provided to ensure usability of the system.Furthermore the system is most useful for semi-interactive querying,depending on users to re?ne their query in a few iterations in order to ensure the desired precision and ef?ciency.
The combination of feature information from multiple media types enabled us to implement simple feature extraction algorithms to solve the problem at hand.Multimodal access also proved to be a very ef?cient way to prune the larger search space,the video stream.Cross-modal correlation is certainly an area to study further.
A direction for future research and extension of system are the development of a domain-independent thesaurus layer between the DBMS and user interfaces.
Another topic to study further is the research and development of caching strategies.Since no caching strategies are applied at this moment,the content representations speci?ed at query time need to be re-calculated for each query issued.A nice feature of our approach is that this can be solved as a database problem,independent of the multimedia retrieval system.Also if Monet is improved in other ways,we gain performance automatically without any direct changes to the multimedia retrieval system itself.
We also think of developing a visual query interface for ordinary users.As the system stands now,it is primarily aimed at expert users with knowledge of video-and audio processing.A visual query could use visual query graphs instead of scripts for query articulation.The query graphs are built from both a selection of available high-level media-type speci?c operations or feature extraction and analysis operations or user de?ned operations.Figure4presents an example of such a graph.We also propose that the construction of user-de?ned nodes may feed the thesaurus mentioned earlier.
10
References
1.T.P.Minka and R.W.Picard.Interactive learning using a“society of models”.Pattern Recognition,
30(4):565–581,April1997.
2.M.G.L.M.van Doorn and A.P.de Vries.The psychology of multimedia databases.In Proceedings of
the5th ACM Digital Libraries Conference(DL’00),pages1–9,San Antonio,Texas,USA,June2000.
3.H.G.P.Bosch,A.P.de Vries,N.Nes,and M.L.Kersten.A case for Image Querying through Image
Spots.In Storage and Retrieval for Media Databases2001,Proc.SPIE volume4315,pages20–30, 2001.
4.M.Flickner,H.Sawhney,and W.Niblack.Query by Image and Video Content:the QBIC System.
IEEE Computer,28(9):23–32,1995.
5.S.F.Chang,W.Chen,H.J.Meng,H.Sundaram,and D.Zhong.VideoQ:An Automated Content Based
Video Search and Retrieval System.In Proceedings of ACM Multimedia,1997.
6.S.F.Chang,W.Chen,and H.Sundaram.Semantic Visual Templates:Linking Features to Semantics.
In Proceedings of the5th IEEE International Conference on Image Processing,volume3,pages531–535,1998.
7.J.Foote.An Overview of Audio Information Retrieval.Multimedia Systems,7(1):2–10,1999.
8. E.Wold,T.Blum,D.Keislar,and J.Wheaton.Content-Based Classi?cation,Search and Retrieval of
Audio.IEEE Multimedia,3(3):27–36,1996.
9. B.Arons.SpeechSkimmer:A System for interactively skimming recorded speech.ACM Transactions
on Computer Human Interaction,4(1):3–38,1997.
10.R.McNab,L.A.Smith,I.H.Witten,C.L.Henderson,and S.J.Cunningham.Towards the Digital Music
Library:Tune Retrieval From Acoustic Input.In Proceedings Digital Libraries’96,pages11–18, 1996.
11.M.S.Hacid,C.Decleir,and J.Kouloumdjian.A Database Approach for Modeling and Querying Video
Data.IEEE Transactions on Knowledge and Data Engineering,12(5),2000.
12.S.Marcus and V.Subrahmanian.Multimedia Database Systems.
https://www.doczj.com/doc/0613524548.html,/projects/hermes/publications/postscripts/mm1.ps,1993.
13.M.Petkovic and W.Jonker.An Overview of Data Models and Query Languages for Video Retrieval.
International Conference on Advances in Infrastructure for Electronic Business,Science and Educa-tion on the Internet,l’Aquila,Italy,July2000(Invited paper),2000.
14.M.L.Kersten and P.Boncz.Monet:An Impressionist Sketch of an Advanced Database System.In
Proceedings Basque International Workshop on Information Technology,1995.
7.Conclusions and Future Research11 15.A.P.de Vries.Challenging Ubiquitous Inverted Files.In Proceedings of the1st DELOS Network of
Excellence Workshop on Information Seeking,Searching and Querying in Digital Libraries,volume 01/W001of ERCIM Workshop Reports,pages71–75,2000.
16.Multimedia Indexing,Retrieval on the basis of Image Processing&Language,and Speech Technol-
ogy.The DRUID Project.http://dis.tpd.tno.nl/druid.
17.NOS.Nederlandse Omroep Stichting.http://www.nos.nl/.
18.MPEG.Coding of Moving Pictures and Associated Audio for Digital Storage Media at up to about
1,5Mbit/s–Part3:Audio.Standard ISO/IEC11172-3:1993.
19.M.A.Broadhead and C.B.Owen.Direct Manipulation of MPEG Compressed Digital Audio.In
Proceedings of Multimedia95(ACM),pages499–509,1995.
20.G.Tzanetakis and P.Cook.Sound Analysis using MPEG Compressed Audio.In Proceedings of the
Internal Conference on Audio,Speech and Signal Processing(ICASSP00),2000.
21.MPEG.Coding of Moving Pictures and Associated Audio for Digital Storage Media at up to about
1,5Mbit/s–Part2:Video.Standard ISO/IEC11172-2:1993.
22.A.Kankanhalli H.Zhang and S.W.Smoliar.Automatic Partitioning of Full-Motion Video.Multimedia
Systems,1:10–28,1993.
https://www.doczj.com/doc/0613524548.html,parison of Automatic Shot Boundary Detection Algorithms.In Image and Video
Processing VII1999,Proc.SPIE3656-29,Jan.1999,1999.
24.U.Gargi and R.Kasturi.An Evaluation of Color Histogram Based Methods in Video Indexing.In
Image Databases and Multi-Media Search,Proc.IDB-MMS’96,Aug1996,1996.
25.E.Scheirer and M.Slaney.Construction and Evaluation of a Robust Multifeature Speech/Music Dis-
criminator.In Proceedings ICASSP1997,pages1331–1334,Munich,Germany,1997.
实验二填料精馏塔等板高度的测定 精馏是化工生产中一个很重要的操作过程。在化工厂和实验室中，精馏操作通常被用来分离均相液体混合物。无论是原料的准备或是产品的精制，往往需要应用精馏操作。精馏塔一般分两大类：填料塔和板式塔。实验室精密分馏采用填料精馏柱。评价精馏设备的分离能力，对于板式塔多采用塔板效率；对于填料塔常以单位高度填料层内所具有的理论板数（亦称理论级数）来表示，或以相当于一层理论板的填料层高度，即所谓的等板高度（亦称理论级当量高度）来表示。本实验是在玻璃精馏柱中，填以玻璃填料，以乙醇一水混合物为工作介质，测定填料的分离能力（即测定玻璃球填料的等板高度）。 一、实验目的及任务 了解填料精馏塔的结构。 掌握精密分馏的操作方法。 测定在一定汽、液相负荷条件下，全回流时的全塔效率及等板高度。 二、实验基本原理 通常采用在全回流条件下，当塔内达到传质、传热平衡时，测定其最小理论板数，进而求出等板高度HETP，来评价精馏柱和填料性能。采用全回流操作条件，达到给定分离目的所需要的理论板数最少，即设备的分离能力达到最大，测定时，免去了回流比等的影响。 精馏过程就是依据混合物中两组分挥发度不同，使未达到平衡的汽、液两相进行充分的接触，最终达到平衡状态。从而使汽相中富含易挥发组分，液相中富含难挥发组分，从而达到分离的目的。在实际操作中，由于接触时间有限且汽、液两相接触不可能十分充足，所以最终的相平衡是不易达到的，相平衡只是过程的极限状态。因此，在开发和设计分离设备时， 应使设备的分离能力尽量提高。对于二元物系可以根据相平衡数据及实验，测定设备的这种分离能力（或效率）。 对于本实验所涉及的乙醇一水二元物系，由相平衡数据在直角坐标纸上绘出相图(x-y相图）。由实验测定全回流条件下的釜液浓度xw及塔顶流出液浓度xD （均为摩尔分数），在相图上图解理论级数NT。根据等板高度的定义，便可以计算出填料的等板高度：
实验五 填料精馏塔理论塔板数的测定 精馏操作是分离、精制化工产品的重要操作。塔的理论塔板数决定混合物 的分离程度，因此，理论板数的实际测定是极其重要的。在实验室内由精馏装 置测取某些数据，通过计算得到该值。这种方法同样可以用于大型装置的理论 板数校核。目前包括实验室在内使用最多的是填料精馏塔。其理论板数与塔结 构、填料形状及尺寸有关。测定时要在固定结构的塔内以一定组成的混合物进 行。 一． 实验目的 1．了解实验室填料塔的结构，学会安装、测试的操作技术。 2．掌握精馏理论，了解精馏操作的影响因素，学会填料精馏塔理论板 数的测定方法 3．掌握高纯度物质的提纯制备方法。 二． 实验原理 精馏是基于汽液平衡理论的一种分离方法。对于双组分理想溶液，平衡时 气相中易挥发组分浓度要比液相中的高；气相冷凝后再次进行汽液平衡，则气 相中易挥发组分浓度又相对提高，此种操作即是平衡蒸馏。经过多次重复的平 衡蒸馏可以使两种组分分离。平衡蒸馏中每次平衡都被看作是一块理论板。精 馏塔就是由许多块理论板组成的，理论板越多，塔的分离效率就越高。板式塔 的理论板数即为该塔的板数，而填料塔的理论板数用当量高度表示。填料精馏 塔的理论板与实际板数未必一致，其中存在塔效率问题。实验室测定填料精馏 塔的理论板数是采用间歇操作，可在回流或非回流条件下进行测定。最常用的 测定方法是在全回流条件下操作，可免去加回流比、馏出速度及其它变量影响，而且试剂能反复使用。不过要在稳定条件下同时测出塔顶、塔釜组成，再由该 组成通过计算或图解法进行求解。具体方法如下： １．计算法 二元组份在塔内具有n 块理论板的第一块板的汽液平衡关系符合平衡方 程式为： 1 11y y -=w w N m x x -+11α （１） y 1——第一块板的气相组成 x w ——塔釜液的组成 m α——全塔(包括再沸器)α（相对挥发度）的几何平均值m α=w p αα N ——理论板数
课程设计说明书 题 目：S H S 20-25型锅炉低硫烟煤 烟 气袋式除尘湿式脱硫系统设计 学生姓名： 周永博 学 院： 能源与动力工程学院 班 级： 环工13-1 指导教师：曹英楠
2016年7 月 1 日 内蒙古工业大学课程设计（论文）任务书 课程名称：大气污染控制工程学院：能源与动力工程学院班级：环工13-1 学生姓名：周永博学号：201320303014 指导教师：曹英楠
技术参数： 锅炉型号：SHS20-25 即，双锅筒横置式室燃炉（煤粉炉），蒸发量20t/h，出口蒸汽压力25MPa 设计耗煤量：2.4t/h 设计煤成分：C Y=75.2% H Y=3% O Y=4% N Y=1% S Y=0.8% A Y=10% W Y=6%； V Y＝18％；属于低硫烟煤 排烟温度：160℃ 空气过剩系数＝1.25 飞灰率＝29％ 烟气在锅炉出口前阻力800Pa 污染物排放按照锅炉大气污染物排放标准中2类区新建排污项目执行。 连接锅炉、净化设备及烟囱等净化系统的管道假设长度150m，90°弯头30个。
参考文献： 《大气污染控制工程》郝吉明、马广大； 《环保设备设计与应用》罗辉..北京.高等教育出版社.1997； 《除尘技术》高香林..华北电力大学.2001.3； 《环保设备?设计?应用》郑铭..北京.化学工业出版社.2001.4； 《火电厂除尘技术》胡志光、胡满银..北京.中国水利水电出版社.2005； 《除尘设备》金国淼..北京.化学工业出版社.2002； 《火力发电厂除尘技术》原永涛..北京.化学工业出版社.2004.10； 《环境保护设备选用手册》鹿政理..北京.化学工业出版社.2002.5； 《工业通风》孙一坚主编..中国建筑工业出版社，1994； 《锅炉及锅炉房设备》奚士光等主编..中国建筑工业出版社，1994； 《除尘设备设计》金国淼主编..上海科学技术出版社，1985； 《环境与工业气体净化技术》. 朱世勇主编.化学工业出版社，2001； 《湿法烟气脱硫系统的安全性及优化》曾庭华，杨华等主编..中国电力出版社；《燃煤烟气脱硫脱硝技术及工程实例》. 钟秦主编.化学工业出版社，2004； 《环保工作者使用手册》. 杨丽芬，李友琥主编.冶金工业出版社，2001； 《工业锅炉房设计手册》航天部第七研究设计院编.中国建筑工业出版社，1986；《火电厂烟气湿法脱硫装置吸收塔的设计》王祖培编.化学工业第二设计院，1995；《大气污染控制工程》. 吴忠标编.科学出版社，2002； 《湿法烟气脱硫吸收塔系统的设计和运行分析》. 曾培华著.电力环境保护，2002。
填料塔工艺尺寸的计算 Document number：NOCG-YUNOO-BUYTT-UU986-1986UT
第三节 填料塔工艺尺寸的计算 填料塔工艺尺寸的计算包括塔径的计算、填料能高度的计算及分段 塔径的计算 1. 空塔气速的确定——泛点气速法 对于散装填料，其泛点率的经验值u/u f =~ 贝恩（Bain ）—霍根（Hougen ）关联式 ，即： 2213lg V F L L u a g ρμερ?? ?????? ? ???????=A-K 14 18 V L V L w w ρρ???? ? ??? ?? （3-1） 即：1124 8 0.23100 1.18363202.59 1.1836lg[ ()1]0.0942 1.759.810.917998.24734.4998.2F u ?????? =- ? ? ??????? 所以：2 F u /（100/3）（）= UF=3.974574742m/s 其中： f u ——泛点气速，m/s; g ——重力加速度，9.81m/s 2 W L =㎏/h W V =7056.6kg/h A=； K=； 取u= F u =2.78220m/s 0.7631D = = = （3-2） 圆整塔径后 D=0.8m 1. 泛点速率校核：2 6000 3.31740.7850.83600 u = =?? m/s 则 F u u 在允许范围内 2. 根据填料规格校核：D/d=800/50=16根据表3-1符合 3. 液体喷淋密度的校核： (1) 填料塔的液体喷淋密度是指单位时间、单位塔截面上液体的喷淋量。
(2) 最小润湿速率是指在塔的截面上，单位长度的填料周边的最小液体体积流量。对于直径不超过75mm 的散装填料，可取最小润湿速率()3min 0.08m /m h w L ?为。 ()32min min 0.081008/w t U L m m h α==?=? （3-3） 22 5358.8957 10.6858min 0.75998.20.7850.8L L w U D ρ= ==>=???? （3-4） 经过以上校验，填料塔直径设计为D=800mm 合理。 填料层高度的计算及分段 *110.049850.75320.03755Y mX ==?= （3-5） *220Y mX == （3-6） 3.2.1 传质单元数的计算 用对数平均推动力法求传质单元数 12 OG M Y Y N Y -= ? （3-7） ()* *1 1 22*11*22 () ln M Y Y Y Y Y Y Y Y Y ---?= -- （3-8） = 0.063830.00063830.03755 0.02627ln 0.0006383 -- = 3.2.2 质单元高度的计算 气相总传质单元高度采用修正的恩田关联式计算： () 0.75 0.10.05 2 0.2 2 21exp 1.45/t c l L t L L V t w l t l L U U U g ασαρσαασαμρ-????????? ? =--?? ? ? ??? ????? ?? ? （3-9） 即：αw/αt =0. 液体质量通量为：L u =WL/××=10666.5918kg/(㎡?h ) 气体质量通量为： V u =60000×=14045.78025kg/(㎡?h)
目录 1.前言 (4) 2.设计任务 (6) 3.设计方案说明 (6) 4.基础物性数据 (6) 5.物料衡算 (6) 6.填料塔的工艺尺寸计算 (8) 7.附属设备的选型及设备 (14) 8.参考文献 (19) 9.后记及其他 (20)
1.前言 填料塔是以塔内的填料作为气液两相间接触构件的传质设备，它是化工类企业中最常用的气液传质设备之一。而塔填料塔内件及工艺流程又是填料塔技术发展的关键。聚丙烯材质填料作为塔填料的重要一类，在化工上应用较为广泛，与其他材质的填料相比，聚丙烯填料具有质轻、价廉、耐蚀、不易破碎及加工方便等优点，但其明显的缺点是表面润湿性能。 1.1填料塔技术 填料塔的塔身是一直立式圆筒，底部装有填料支承板，填料以乱堆或整砌的方式放置在支承板上。填料的上方安装填料压板，以防被上升气流吹动。液体从塔顶经液体分布器喷淋到填料上，并沿填料表面流下。气体从塔底送入，经气体分布装置（小直径塔一般不设气体分布装置）分布后，与液体呈逆流连续通过填料层的空隙，在填料表面上，气液两相密切接触进行传质。填料塔属于连续接触式气液传质设备，两相组成沿塔高连续变化，在正常操作状态下，气相为连续相，液相为分散相。 当液体沿填料层向下流动时，有逐渐向塔壁集中的趋势，使得塔壁附近的液流量逐渐增大，这种现象称为壁流。壁流效应造成气液两相在填料层中分布不均，从而使传质效率下降。因此，当填料层较高时，需要进行分段，中间设置再分布装置。液体再分布装置包括液体收集器和液体再分布器两部分，上层填料流下的液体经液体收集器收集后，送到液体再分布器，经重新分布后喷淋到下层填料上。 填料塔具有生产能力大，分离效率高，压降小，持液量小，操作弹性大等优点。填料塔也有一些不足之处，如填料造价高；当液体负荷较小时不能有效地润湿填料表面，使传质效率降低；不能直接用于有悬浮物或容易聚合的物料；对侧线进料和出料等复杂精馏不太适合等。 1.2 填料的类型 填料的种类很多，根据装填方式的不同，可分为散装填料和规整填料。 散装填料是一个个具有一定几何形状和尺寸的颗粒体，一般以随机的方式堆积在塔内，又称为乱堆填料或颗粒填料。散装填料根据结构特点不同，又可分为环形填料、鞍形填料、环鞍形填料及球形填料等。
填料塔计算部分 This manuscript was revised by the office on December 10, 2020.
二 基础物性参数的确定 1 液相物性数据 对于低浓度吸收过程，溶液的物性数据可近似取纯水的物性数据。由手册查得，2 气相物性参数 设计压力： ,温度：20C ? 氨气在水中的扩散系数：92621.7610/ 6.33610/L D cm s m h --=?=? 氨气在空气中的扩散系数： 查表得，氨气在0°C ，在空气中的扩散系数为 2/cm s ， 根据关系式换算出20C ?时的空气中的扩散系数： 3 32 2 00022293.150.171273.150.189/0.06804/V P T D D P T cm s m h ?????? ==?? ? ? ??????? == 混合气体的平均摩尔质量为 m i 0.05170.982929.27V i M y M ==?+?=∑ 混合气体的平均密度为 3m 101.329.27 1.2178.314293.15 V Vm PM kg m RT ρ?===? 混合气体的粘度可近似取空气的粘度，查手册得20C ?空气粘度为 51.81100.065()V Pa s kg m h μ-=??=? 3 气液相平衡数据
由手册查得，常压下20C ?时，氨气在水中的亨利系数 76.3a E kP = 相平衡常数 76.30.7532101.3 E m P === 溶解度系数 3s 998.2 0.726076.318.02 L H kmol kPa m EM ρ= = =?? 4 物料衡算 进塔气相摩尔比 1= 110.05 0.05263110.05 y Y y ==-- 出塔气相摩尔比 321(1)0.05263(10.98) 1.05310A Y Y ?-=-=-=? 混合气体流量 330.1013(273.1520) 16.10100.1013273.15 V N Q Q m h ? ?+==?? 惰性气体摩尔流量 273.15(10.05)636.1622.4273.1520 V Q V kmol h =?-=+ 该吸收过程属低浓度吸收，平衡关系为直线，最小液气比可按下式计算： 1212 L Y Y V Y m X -??= ? -?? 对于纯溶剂吸收过程，进塔液相组成 20X = min 0.052630.0010530.73810.052630.7532L V -??== ??? 取操作液气比为 min 1.4L L V V ?? = ??? 1.40.7381 1.0333L V =?= 1.0333636.16657.34L kmol h =?= 1212()636.16(0.052630.001053) 0.0499657.34 V Y Y X X L -?-=+==
填料塔计算和设计
填料塔计算和设计 Pleasure Group Office【T985AB-B866SYT-B182C-BS682T-STT18】
填料塔设计 2012-11-20 一、填料塔结构 填料塔是以塔内装有大量的填料为相间接触构件的气液传质设备。填料塔的塔身是一直立式圆筒，底部装有填料支承板，填料以乱堆或整砌的方式放置在支承板上。在填料的上方安装填料压板，以限制填料随上升气流的运动。液体从塔顶加入，经液体分布器喷淋到填料上，并沿填料表面流下。气体从塔底送入，经气体分布装置（小直径塔一般不设置）分布后，与液体呈逆流接触连续通过填料层空隙，在填料表面气液两相密切接触进行传质。填料塔属于连续接触式的气液传质设备，正常操作状态下，气相为连续相，液相为分散相。二、填料的类型及性能评价 填料是填料塔的核心构件，它提供了气液两相接触传质的相界面，是决定填料塔性能的主要因素。填料的种类很多，根据装填方式的不同，可分为散装填料和规整填料两大类。散装填料根据结构特点不同，分为环形填料、鞍形填料、环鞍形填料等；规整填料按其几何结构可分为格栅填料、波纹填料、脉冲填料等，目前工业上使用最为广泛的是波纹填料，分为板波纹填料和网波纹填料； 填料的几何特性是评价填料性能的基本参数，主要包括比表面积、空隙率、填料因子等。1.比表面积：单位体积填料层的填料表面积，其值越大，所提供的气液传质面积越大，性能越优； 2.空隙率：单位体积填料层的空隙体积；空隙率越大，气体通过的能力大且压降低；
3.填料因子：填料的比表面积与空隙率三次方的比值，它表示填料的流体力学性能，其值越小，表面流体阻力越小。 三、填料塔设计基本步骤 1.根据给定的设计条件，合理地选择填料； 2.根据给定的设计任务，计算塔径、填料层高度等工艺尺寸； 3.计算填料层的压降； 4.进行填料塔的结构设计，结构设计包括塔体设计及塔内件设计两部分。 四、填料塔设计 1.填料的选择 填料应根据分离工艺要求进行选择，对填料的品种、规格和材质进行综合考虑。应尽量选用技术资料齐备，适用性能成熟的新型填料。对性能相近的填料，应根据它的特点进行技术经济评价，使所选用的填料既能满足生产要求，又能使设备的投资和操作费最低。 （1）填料种类的选择 填料的传质效率要高：传质效率即分离效率，一般以每个理论级当量填料层高度表示，即HETP值； 填料的通量要大：在同样的液体负荷下，在保证具有较高传质效率的前提下，应选择具有较高泛点气速或气相动能因子的填料； 填料层的压降要低：填料层压降越低，塔的动力消耗越低，操作费越小；对热敏性物系尤为重要；
1.填料塔的一般结构 填料塔可用于吸收气体等。填料塔的主要组件是：流体分配器，填料板或床限制板，填料，填料支架，液体收集器，液体再分配器等。 2.填料塔的设计步骤 （1）确定气液负荷，气液物理参数和特性，根据工艺要求确定出气口上述参数（2）填料的正确选择对塔的经济效果有重要影响。对于给定的设计条件，有多种填充物可供选择。因此，有必要对各种填料进行综合比较，限制床层，以选择理想的填料。 （3）塔径的计算：根据填料特性数据，系统物理参数和液气比计算出驱替速度，再乘以适当的系数，得出集液器设计的空塔气速度，以计算塔径。;或者直接使用从经验中获得的气体动能因子的设计值来计算塔的直径。 （4）填充层的总高度通过传质单位高度法或等板高度法算出。
（5）计算填料层的压降。如果压降超过极限值，则应调整填料的类型和尺寸或降低工作气体的速度，然后再重复计算直至满足条件。 （6）为了确保填料塔的预期性能，填料塔的其他内部组件（分配器，填料支座，再分配器，填料限位板等）必须具有适当的设计和结构。结构设计包括两部分：塔身设计和塔内构件设计。填料塔的内部组件包括：液体分配装置，液体再分配装置，填料支撑装置，填料压板或床限制板等。这些内部构件的合理设计是确保正常运行和预期性能的重要条件。 废气处理设备 第六章小型吸收塔的设计32参考文献33设计师：武汉工程大学环境工程学院08级环境工程去除工艺气体中更多的有害成分以净化气体以进一步处理或去除工业废气中的更多有害物质，以免造成空气污染。1.2吸收塔的应用塔式设备是气液传质设备，广泛用于炼油，化工，石家庄汕头化工等生产。根部列车塔中气液接触部分的结构类型可分为板式塔和填料塔。根据气体和液体的接触方式的不同，吸收设备可分为两类：阶
填料塔计算和设计文件编码（008-TTIG-UTITD-GKBTT-PUUTI-WYTUI-8256）
填料塔设计 2012-11-20 一、填料塔结构 填料塔是以塔内装有大量的填料为相间接触构件的气液传质设备。填料塔的塔身是一直立式圆筒，底部装有填料支承板，填料以乱堆或整砌的方式放置在支承板上。在填料的上方安装填料压板，以限制填料随上升气流的运动。液体从塔顶加入，经液体分布器喷淋到填料上，并沿填料表面流下。气体从塔底送入，经气体分布装置（小直径塔一般不设置）分布后，与液体呈逆流接触连续通过填料层空隙，在填料表面气液两相密切接触进行传质。填料塔属于连续接触式的气液传质设备，正常操作状态下，气相为连续相，液相为分散相。 二、填料的类型及性能评价 填料是填料塔的核心构件，它提供了气液两相接触传质的相界面，是决定填料塔性能的主要因素。填料的种类很多，根据装填方式的不同，可分为散装填料和规整填料两大类。散装填料根据结构特点不同，分为环形填料、鞍形填料、环鞍形填料等；规整填料按其几何结构可分为格栅填料、波纹填料、脉冲填料等，目前工业上使用最为广泛的是波纹填料，分为板波纹填料和网波纹填料； 填料的几何特性是评价填料性能的基本参数，主要包括比表面积、空隙率、填料因子等。
1.比表面积：单位体积填料层的填料表面积，其值越大，所提供的气液传质面积越大，性能越优； 2.空隙率：单位体积填料层的空隙体积；空隙率越大，气体通过的能力大且压降低； 3.填料因子：填料的比表面积与空隙率三次方的比值，它表示填料的流体力学性能，其值越小，表面流体阻力越小。 三、填料塔设计基本步骤 1.根据给定的设计条件，合理地选择填料； 2.根据给定的设计任务，计算塔径、填料层高度等工艺尺寸； 3.计算填料层的压降； 4.进行填料塔的结构设计，结构设计包括塔体设计及塔内件设计两部分。？ 四、填料塔设计 1.填料的选择 填料应根据分离工艺要求进行选择，对填料的品种、规格和材质进行综合考虑。应尽量选用技术资料齐备，适用性能成熟的新型填料。对性能相近的填料，应根据
一、 设计方案的确定 (一) 操作条件的确定 1．1吸收剂的选择 1．2装置流程的确定 1．3填料的类型与选择 1．4操作温度与压力的确定 45℃ 常压 （二）填料吸收塔的工艺尺寸的计算 2．1基础物性数据 ①液相物性数据 对于低浓度吸收过程，溶液的物性数据可近似取质量分数为30%MEA 的物性数据 7.熔 根据上式计算如下： 混合密度是：1013.865KG/M3 混合粘度0.001288 Pa ·s 暂取CO2在水中的扩散系数 表面张力б=72.6dyn/cm=940896kg/h 3 ②气相物性数据 混合气体的平均摩尔质量为 M vm = y i M i =0.133*44+0.0381*64+0.7162*14+0.00005*96+0.1125*18 =20.347 混合气体的平均密度ρvm = =??=301 314.805.333.101RT PMvm 101.6*20.347/（8.314*323）=0.769kg/m 3
混合气体粘度近似取空气粘度，手册28℃空气粘度为 μV =1.78×10-5Pa ·s=0.064kg/(m?h) 查手册得CO2在空气中的扩散系数为 D V =1.8×10-5m 2/s=0.065m 2 /h 由文献时CO 2在MEA 中的亨利常数： 在水中亨利系数E=2.6?105kPa 相平衡常数为m=1.25596 .101106.25 =?=P E 溶解度系数为H= )/(1013.218 106.22.997345kPa m kmol E M s ??=??=-ρ 2.2物料衡算 进塔气相摩尔比为Y1=0.133/（1-0.133）= 0.153403 出塔气相摩尔比为Y2= 0.153403×0.05=0.00767 进塔惰性气相流量为V=992.1mol/s=275.58kmol/h 该吸收过程为低浓度吸收，平衡关系为直线，最小液气比按下式计算，即 2121min /X m Y Y Y ）V L （ --= 对于纯溶剂吸收过程，进塔液组成为X2=0 2121min /X m Y Y Y ）V L （ --==（0.153403-0.00767）/（0.1534/1.78）=1.78 取操作液气比（？）为L/V=1.5L/V=1.5×1.78=2.67 L=2.67×275.58=735.7986kmol/h ∵V(Y1-Y2)=L(X1-X2) ∴X1=0.054581 ①塔径计算 采用Eckert 通用关联图计算泛点气速 气相质量流量为 W V =13.74kg/s=49464kg/h 液相质量流量计算 即W L =735.7986×（0.7*18+0.3*54）=21190.99968kg/h Eckert 通用关联图横坐标为 0.011799 查埃克特通用关联图得226.02.0=??L L V F F g u μρρ?φ（查表相差不多） 查表（散装填料泛点填料因子平均值）得1260-=m F φ Uf=3.964272m/s 取u=0.8u F =0.8×3.352=2.6816m/s
学校代码： 10128 学号： 201320303014 课程设计说明书 题目：S H S20-25型锅炉低硫烟煤烟 气袋式除尘湿式脱硫系统设计学生：周永博 学院：能源与动力工程学院 班级：环工13-1 指导教师：英楠
2016年 7 月 1 日 工业大学课程设计（论文）任务书 课程名称：大气污染控制工程学院：能源与动力工程学院班级：环工13-1 学生：周永博学号： 4 指导教师：英楠
技术参数： 锅炉型号：SHS20-25 即，双锅筒横置式室燃炉（煤粉炉），蒸发量20t/h，出口蒸汽压力25MPa 设计耗煤量：2.4t/h 设计煤成分：C Y=75.2% H Y=3% O Y=4% N Y=1% S Y=0.8% A Y=10% W Y=6%； V Y＝18％；属于低硫烟煤 排烟温度：160℃ 空气过剩系数＝1.25 飞灰率＝29％ 烟气在锅炉出口前阻力800Pa 污染物排放按照锅炉大气污染物排放标准中2类区新建排污项目执行。 连接锅炉、净化设备及烟囱等净化系统的管道假设长度150m，90°弯头30个。
参考文献： 《大气污染控制工程》郝吉明、马广大； 《环保设备设计与应用》罗辉...高等教育.1997； 《除尘技术》高香林..华北电力大学.2001.3； 《环保设备?设计?应用》铭...化学工业.2001.4； 《火电厂除尘技术》胡志光、胡满银...中国水利水电.2005； 《除尘设备》金国淼...化学工业.2002； 《火力发电厂除尘技术》原永涛...化学工业.2004.10； 《环境保护设备选用手册》鹿政理...化学工业.2002.5； 《工业通风》一坚主编..中国建筑工业，1994； 《锅炉及锅炉房设备》奚士光等主编..中国建筑工业，1994； 《除尘设备设计》金国淼主编..科学技术，1985； 《环境与工业气体净化技术》. 朱世勇主编.化学工业，2001； 《湿法烟气脱硫系统的安全性及优化》曾庭华，华等主编..中国电力； 《燃煤烟气脱硫脱硝技术及工程实例》. 钟主编.化学工业，2004； 《环保工作者使用手册》. 丽芬，友琥主编.冶金工业，2001； 《工业锅炉房设计手册》航天部第七研究编.中国建筑工业，1986； 《火电厂烟气湿法脱硫装置吸收塔的设计》王祖培编.化学工业第二，1995； 《大气污染控制工程》. 标编.科学，2002； 《湿法烟气脱硫吸收塔系统的设计和运行分析》. 曾培华著.电力环境保护，2002。
目录 一．设计任务书 (3) 1.设计目的 (3) 2.设计任务 (3) 3.设计内容和要求 (3) 二．设计资料 (4) 1.工艺流程 (4) 2.进气参数 (4) 3.吸收液参数 (4) 4.操作条件 (5) 5.填料性能 (5) 三．设计计算书 (6) 1．填料塔主体的计算 (6) 1.1吸收剂用量的计算 (6) 1.2塔径的计算 (7) 1.3填料层高度的计算 (10) 1.4.填料塔压降的计算 (14) 2.填料塔附属结构的类型与设计 (15) 2.1支承板 (16) 2.2填料压紧装置 (16) 2.3液体分布器装置 (16) 2.4除雾装置 (17) 2.5气体分布装置 (17) 2.6排液装置 (18)
2.7防腐蚀设计 (18) 2.8气体进料管 (18) 2.9液体进料管： (19) 2.10封头的选择 (19) 2.11总塔高计算 (20) 3.填料塔设计参数汇总 (21) 四．填料塔装配图（见附录） (22) 五．总结 (22) 六．参考文献 (23) 附录 (23)
前言 世界卫生组织和联合国环境组织发表的一份报告说：“空气污染已成为全世界城市居民生活中一个无法逃避的现实。”如果人类生活在污染十分严重的空气里，那就将在几分钟内全部死亡。工业文明和城市发展，在为人类创造巨大财富的同时，也把数十亿吨计的废气和废物排入大气之中，人类赖以生存的大气圈却成了空中垃圾库和毒气库。因此，大气中的有害气体和污染物达到一定浓度时，就会对人类和环境带来巨大灾难，对有害气体的控制更必不可少。 一．设计任务书 1.设计目的 通过对气态污染物净化系统的工艺设计，初步掌握气态污染物净化系统设计的基本方法。培养学生利用所学理论知识，综合分析问题和解决实际问题的能力、绘图能力、以及正确使用设计手册和相关资料的能力。 2.设计任务 试设计一个填料塔，常压，逆流操作，操作温度为25℃，以清水为吸收剂，吸收脱除混合气体中的NH ，气体处理量为1500m3/h，其中含氨1.9%（体积分数）， 3 要求吸收率达到99%，相平衡常数m=0.95。 3.设计内容和要求 1）研究分析资料。 2）净化设备的计算，包括计算吸收塔的物料衡算、吸收塔的工艺尺寸计算、填料层压降的计算及校核计算。 3）附属设备的设计等。 4）编写设计计算书。设计计算书的内容应按要求编写，即包括与设计有关的阐述、说明及计算。要求内容完整，叙述简明，层次清楚，计算过程详细、准确，书写工整，装订成册。设计计算书应包括目录、前言、正文及参考文献等，格式参照学校要求。 5）设计图纸。包括填料塔剖面结构图、工艺流程图。应按比例绘制，标出设备、
第三节 填料塔工艺尺寸的计算 填料塔工艺尺寸的计算包括塔径的计算、填料能高度的计算及分段 塔径的计算 1. 空塔气速的确定——泛点气速法 对于散装填料，其泛点率的经验值u/u f =~ 贝恩（Bain ）—霍根（Hougen ）关联式 ，即： 221 3lg V F L L u a g ρμερ?? ?????? ? ?????? ?=A-K 14 18 V L V L w w ρρ???? ? ??? ?? （3-1） 即：1 124 8 0.23100 1.18363202.59 1.1836lg[ ()1]0.0942 1.759.810.917998.24734.4998.2F u ?????? =- ? ? ??????? 所以：2 F u /（100/3）（）= UF=3.974574742m/s 其中： f u ——泛点气速，m/s; g ——重力加速度，9.81m/s 2 23t m /m α--填料总比表面积， 33m /m ε--填料层空隙率 33 V 998.2/1.1836kg /m l kg m ρρ==液相密度。气相密度 W L =㎏/h W V =7056.6kg/h A=； K=； 取u= F u =2.78220m/s 0.7631D = = = （3-2） 圆整塔径后 D=0.8m 1. 泛点速率校核：26000 3.31740.7850.83600 u = =?? m/s
3.31740.83463.9746 F u u == 则 F u u 在允许范围内 2. 根据填料规格校核：D/d=800/50=16根据表3-1符合 3. 液体喷淋密度的校核： (1) 填料塔的液体喷淋密度是指单位时间、单位塔截面上液体的喷淋量。 (2) 最小润湿速率是指在塔的截面上，单位长度的填料周边的最小液体体积流量。对于直径不超过75mm 的散装填料，可取最小润湿速率 ()3min 0.08m /m h w L ?为。 ()32min min 0.081008/w t U L m m h α==?=? （3-3） 22 5358.8957 10.6858min 0.75998.20.7850.8L L w U D ρ= ==>=???? （3-4） 经过以上校验，填料塔直径设计为D=800mm 合理。 填料层高度的计算及分段 *110.049850.75320.03755Y mX ==?= （3-5） *220Y mX == （3-6） 3.2.1 传质单元数的计算 用对数平均推动力法求传质单元数 12 OG M Y Y N Y -= ? （3-7） ()**1 1 2 2* 11* 22() ln M Y Y Y Y Y Y Y Y Y ---?= -- （3-8） = 0.063830.00063830.03755 0.02627ln 0.0006383 -- =
( 目录 一．设计任务书 (2) 1.设计目的 (2) 2.设计任务 (2) 3.设计内容和要求 (2) 二．设计资料 (3) 1.工艺流程 (3) 2.进气参数 (3) ) 3.吸收液参数 (3) 4.操作条件 (3) 5.填料性能 (4) 三．设计计算书 (5) 1．填料塔主体的计算 (5) 吸收剂用量的计算 (5) 塔径的计算 (6) 填料层高度的计算 (8) ' .填料塔压降的计算 (12) 2.填料塔附属结构的类型与设计 (13) 支承板 (13) 填料压紧装置 (13)
液体分布器装置 (13) 除雾装置 (14) 气体分布装置 (14) 排液装置 (15) ] 防腐蚀设计 (15) 气体进料管 (15) 液体进料管： (16) 封头的选择 (16) 总塔高计算 (16) 3.填料塔设计参数汇总 (18) 四．填料塔装配图（见附录） (19) 五．总结 (19) ！ 六．参考文献 (19) 附录 (20)
前言 世界卫生组织和联合国环境组织发表的一份报告说：“空气污染已成为全世界城市居民生活中一个无法逃避的现实。”如果人类生活在污染十分严重的空气里，那就将在几分钟内全部死亡。工业文明和城市发展，在为人类创造巨大财富的同时，也把数十亿吨计的废气和废物排入大气之中，人类赖以生存的大气圈却成了空中垃圾库和毒气库。因此，大气中的有害气体和污染物达到一定浓度时，就会对人类和环境带来巨大灾难，对有害气体的控制更必不可少。 一．设计任务书 1.设计目的 ' 通过对气态污染物净化系统的工艺设计，初步掌握气态污染物净化系统设计的基本方法。培养学生利用所学理论知识，综合分析问题和解决实际问题的能力、绘图能力、以及正确使用设计手册和相关资料的能力。 2.设计任务 试设计一个填料塔，常压，逆流操作，操作温度为25℃，以清水为吸收剂，吸收脱除混合气体中的NH ，气体处理量为1500m3/h，其中含氨%（体积分数）， 3 要求吸收率达到99%，相平衡常数m=。 3.设计内容和要求 1）研究分析资料。 2）净化设备的计算，包括计算吸收塔的物料衡算、吸收塔的工艺尺寸计算、填料层压降的计算及校核计算。 3）附属设备的设计等。 4）编写设计计算书。设计计算书的内容应按要求编写，即包括与设计有关的阐述、说明及计算。要求内容完整，叙述简明，层次清楚，计算过程详细、准确，书写工整，装订成册。设计计算书应包括目录、前言、正文及参考文献等，格式参照学校要求。
等板高度 英文名称：height equivalent of theoretical plate 简称：HETP HETP又称理论板当量高度。指填料层或喷淋塔固体颗粒移动床的一段高度，其效果与一层理论塔板或一理论级相等。等板高度乘以分离所要求的理论板数即为所需的填料总高，或喷淋塔、移动床的有效高。等板高度的值愈小，则塔内这一段的传质效果愈佳。 影响HETP的因素有：系统的物性、几何因素及操作条件。在应用时宜采取最接近客观情况的实测值。 (1)用于精馏时，填料直径：d=25mm时，HETP为0.46m；d=38mm 时，HETP为0.66m；d=50mm时，HETP为0.9m。 (2)用于吸收时，HETP为1.5～1.8m。 (3)用于小塔[塔径<0.6m]时，HETP等于塔径。 (4)用于真空操作时，HETP在上述数据加0.1。 此外也可用一些经验式作估算。 填料 半软性填料
填充塔内的惰性固体物料,例如鲍尔环和拉西环等，其作用是增大气-液的接触面,使其相互强烈混合。在化工产品中，填料又称填充剂，是指用以改善加工性能、制品力学性能并（或）降低成本的固体物料。 目录 填料 tiánliào [filling，stuffing] 可作填充物的东西 填料的概述 填料[1]也称作填充剂、增量剂。某些填料同时又是体质颜料。微纽的填料具有良好的遮盖力，常用于涂料行业。 填料可用于多种聚氨酯制品，例如聚氨酯涂举}、密封胶：聚氨酯浆料、特殊弹性体i聚氨酯泡沫塑料。三聚氰胺植物纤维聚合，皂参多元醇等，有机填料可用于聚氨酯泡沫塑料;碳酸钙高岭土(陶土、·瓷土)，分子筛粉末滑石粉硅灰石滟技钛白粉j重晶石粉(硫酸钡)’等微细无机粉末二般可用作聚氨酯密封胶0聚氨酯软泡聚氨酯弹性体，胶黏剂。聚氨酯涂料等的填料。
填料塔设计 2012-11-20 一、填料塔结构 填料塔是以塔内装有大量的填料为相间接触构件的气液传质设备。填料塔的塔身是一直立式圆筒，底部装有填料支承板，填料以乱堆或整砌的方式放置在支承板上。在填料的上方安装填料压板，以限制填料随上升气流的运动。液体从塔顶加入，经液体分布器喷淋到填料上，并沿填料表面流下。气体从塔底送入，经气体分布装置（小直径塔一般不设置）分布后，与液体呈逆流接触连续通过填料层空隙，在填料表面气液两相密切接触进行传质。填料塔属于连续接触式的气液传质设备，正常操作状态下，气相为连续相，液相为分散相。 二、填料的类型及性能评价 填料是填料塔的核心构件，它提供了气液两相接触传质的相界面，是决定填料塔性能的主要因素。填料的种类很多，根据装填方式的不同，可分为散装填料和规整填料两大类。散装填料根据结构特点不同，分为环形填料、鞍形填料、环鞍形填料等；规整填料按其几何结构可分为格栅填料、波纹填料、脉冲填料等，目前工业上使用最为广泛的是波纹填料，分为板波纹填料和网波纹填料； 填料的几何特性是评价填料性能的基本参数，主要包括比表面积、空隙率、填料因子等。
1.比表面积：单位体积填料层的填料表面积，其值越大，所提供的气液传质面积越大，性能越优； 2.空隙率：单位体积填料层的空隙体积；空隙率越大，气体通过的能力大且压降低； 3.填料因子：填料的比表面积与空隙率三次方的比值，它表示填料的流体力学性能，其值越小，表面流体阻力越小。 三、填料塔设计基本步骤 1.根据给定的设计条件，合理地选择填料； 2.根据给定的设计任务，计算塔径、填料层高度等工艺尺寸； 3.计算填料层的压降； 4.进行填料塔的结构设计，结构设计包括塔体设计及塔内件设计两部分。 四、填料塔设计 1.填料的选择 填料应根据分离工艺要求进行选择，对填料的品种、规格和材质进行综合考虑。应尽量选用技术资料齐备，适用性能成熟的新型填料。对性能相近的填料，应根据它的特点进行技术经济评价，使所选用的填料既能满足生产要求，又能使设备的投资和操作费最低。 （1）填料种类的选择 填料的传质效率要高：传质效率即分离效率，一般以每个理论级当量填料层高度表示，即HETP值；
一、设计方案的确定 (一) 操作条件的确定 1．1吸收剂的选择 1．2装置流程的确定 1．3填料的类型与选择 1．4操作温度与压力的确定 45℃常压 （二）填料吸收塔的工艺尺寸的计算 2．1基础物性数据 ①液相物性数据 对于低浓度吸收过程，溶液的物性数据可近似取质量分数为30%MEA 的物性数据
7.熔 根据上式计算如下： 混合密度是：1013.865KG/M3 混合粘度0.001288 Pa ·s 暂取CO2在水中的扩散系数 表面张力б=72.6dyn/cm=940896kg/h 3 ②气相物性数据 混合气体的平均摩尔质量为 M vm = y i M i =0.133*44+0.0381*64+0.7162*14+0.00005*96+0.1125*18 =20.347 混合气体的平均密度ρ vm = =??=301 314.805 .333.101RT PMvm 101.6*20.347/（8.314*323）=0.769kg/m 3 混合气体粘度近似取空气粘度，手册28℃空气粘度为
μV =1.78×10-5Pa ·s=0.064kg/(m ?h) 查手册得CO2在空气中的扩散系数为 D V =1.8×10-5m 2/s=0.065m 2/h 由文献时CO 2在MEA 中的亨利常数： 在水中亨利系数E=2.6?105kPa 相平衡常数为m=1.25596 .101106.25 =?= P E 溶解度系数为H=)/(1013.218 106.22.9973 45 kPa m kmol E M s ??=??= -ρ 2.2物料衡算 进塔气相摩尔比为Y1=0.1（1-0.133）= 0.153403 出塔气相摩尔比为Y2= 0.153403×0.05=0.00767 进塔惰性气相流量为V=992.1mol/s=275.58kmol/h 该吸收过程为低浓度吸收，平衡关系为直线，最小液气比按下式 计算，即 2121min /X m Y Y Y ）V L （ --= 对于纯溶剂吸收过程，进塔液组成为X2=0 2121min /X m Y Y Y ）V L （ --==（0.153403-0.00767）/（0.1534/1.78）=1.78 取操作液气比（？）为L/V=1.5L/V=1.5×1.78=2.67 L=2.67×275.58=735.7986kmol/h ∵V(Y1-Y2)=L(X1-X2) ∴X1=0.054581
《化工设备机械基础》 填料塔设计 学院：合肥学院 班级：09化工（3）班 学号：0903023005 0903023001 0903023002 0903023013 姓名：王雷唐显泽王辉伍石
填料塔设计 前言:填料吸收塔简介 在化学工业中，吸收操作广泛应用于石油炼制，石油化工中分离气体混合物，原料气的精制及从废气回收有用组分或去除有害组分等。吸收操作中以填料吸收塔生产能力大，分离效率高，压力降小，操作弹性大和持液量小等优点而被广泛应用。目前国内对填料吸收塔设计大部分是经验设计方法，该方法是在给定生产任务的条件下，由经验确定出一个液气比的值，然后手算出吸收塔的有关设计参数。该设计手段落后，没有考虑经济技术指标，不符合工厂实际生产中成本最低要求，故提出了填料吸收塔的优化设计方法。 下面简要介绍一下填料塔的有关内容。 填料塔属于连续接触式气液传质设备，两相组成沿塔高连续变化，在正常操作状态下，气相为连续相，液相为分散相。填料塔以塔内的填料作为气液两相间接触构件的传质设备。填料塔的塔身是一直立式圆筒，底部装有填料支承板，填料以乱堆或整砌的方式放置在支承板上。填料的上方安装填料压板，以防被上升气流吹动。液体从塔顶经液体分布器喷淋到填料上，并沿填料表面流下。气体从塔底送入，经气体分布装置分布后，与液体呈逆流连续通过填料层的空隙，在填料表面上，气液两相密切接触进行传质。 与板式塔相比，在填料塔中进行的传质过程，其特点是气液连续接触，而传质的好坏与填料密切相关。填料提供了塔内的气液两相接触面积。填料塔的流体力学性能，传质速率等与填料的材质，几何形状密切相关，所以长期以来人们十分注中填料的性能和新型填料的开发，使得填料塔在化工生产中应用更加广泛。填料塔具有生产能力大，分离效率高，压降小，持液量小，操作弹性大等优点。填料塔还有以下特点： 1.当塔径不是很大时，填料塔因为结构简单而造价便宜。 2.对于易起泡物系，填料塔更适合，因填料对气泡有限制和破碎作用。 3.对于腐蚀性物系，填料塔更适合，因为可以采用瓷质填料。 4.对于热敏性物系宜采用填料塔，因为填料塔的持液量比板式塔少，物料在塔内的停留时间短。填料塔的压强降比板式塔小，因而对真空操作更有利。 填料塔也有一些不足之处，如填料造价高；当液体负荷较小时不能有效地润湿填料表面，使传质效率降低；不能直接用于有悬浮物或容易聚合的物料；对侧线进料和出料等复杂精馏不太适合等。 填料塔的类型很多，其设计的原则大体相同，一般来说，填料塔的设计步骤如下： ①根据设计任务和工艺要求，确定设计方案； ②根据设计任务和工艺要求，合理地选择填料； ③确定塔径、填料层高度等工艺尺寸；