On-line Data Analysis and Control Systems for Experiments, Lecture notes of Physics

Download On-line Data Analysis and Control Systems for Experiments and more Lecture notes Physics in PDF only on Docsity! SLAC m-127 August 1965 W. F. Miller Stanford Linear Accelerator Center Stanford University, Stanford, California IBM Scientific Symposium on Man-Machine Communication at The Thomas J. Watson Research Center Yorktown Heights, New York WY 5, 1965 Contents I. ~ODUCTION.....................l II. CURREDi'I' SYSTEMS ................... 3 III. GRAPHIC DATAPROCESSING ............... 5 Iv. CONCLUSIONS . . . . . . . . . . . . . . . . . . . . .lO REFERENCE8 . . . . . . . . . . . . . . . . . . ...12 CAPTIONS FOR FIGURES . . . . . . . . . . . . . . . . .14 * Work supported by the U.S. Atomic Energy Commission. -- I. INTRODUCTION Most of the work presented in this Symposium has been concerned with problems that arise in the developent of systems that permit a man rapid access to computational results. This paper is concerned with problems that arise in the developent of systems that are em- ployed fbr data ab1ysi.s ani conl:01 ir cc:lplex 2mriments. Such systems might be represented schematically by the block diagram in Figure 1. Such a diagram has the same general structure as a guid- ance system, and, indeed, it should, since its function is to guide an experiment. Such systems, of course, g ive rise to many of the same problems already discussed; but, on the oib-?r hand, they introduce some additional ones. The motivation for on-line data analysis and control systems for experiments is very much the same as the motivation for rapid access to computational results. Both developments are motivated in prt by the hypothesis that rapid feedback is essential to learning. It is clearly hoped, and by now it may already have happened, that a researcher sitting at his console will make discoveries by seeing his results feed back practically instantly. In the experimental systems, it is also expected that a researcher, by having his data quickly analyzed and by having intimate control over his experiment, can make discoveries and direct the course of the experiment in a more profitable way. Let us look a moment at the scope of the problem. As physics experiments probe deeper and deeper into the fundamental constituents of matter, the experiments become much more complex and the interpre- tation of the data becomes more subtle. A single experiment involving a high energy particle accelerator and a piece of detection apparatus, 4 have the generality that would be required of such a system in high energy particle physics. Nonetheless, scme quite valuable experience was gained in programmed control of experiments. Now, how do we stand in the very complex experiments such as we encounter in high energy particle physics? We i-re c-srtainiy making progress in all areas indicated in Figure 1. At present, there are no high energy particle accelerators operating under computer control. [By computer, here, Imean interr&lystored program computer]. At two Centers, however, computer controls are being developed. A digi- tal control system, employing a CDC 924, has been developed for the Zero Gradient Synchrotron (ZGS) at the Argonne National Laboratory. It is expected that this system will soon be "hooked up" and that the ZGS will operate under its control. At the Stanford Linear Accelerator Center (SLAC) a computer control system is being developed for the beam switchyard. The beam switchyard will employ bending magnets to _ deflect the electron beam from the accelerator into different experi- mental areas. The problems represented by box 3 are in somewhat better shape. Several on-line data analyzers have been employed in low energy nuclear physics [References 5, 6, 7, and 8].and one system has been developed in high energy physics with great success [Reference 91. At Stanford we are also developing a system similar to that described in Reference 9. This system is intended to analyze on-line the data generated by the 20 BeV/C Magnetic Spectrometer. We are also developing a computer system that will permit on-line analysis of graphic data of the kind we shall get from filmless spark chambers. There are two large general problems that have to be dealt with: (1) th e control programs to permit processing of data from 5 devices with high burst rates, and (2) the data analysis techniques for handling the gra-@ic data. Both of these areas are getting a great deal of attention and are important to the develoment of in- tegrated systems. I shall give below a discussion of the type of programs developed to handle the data analysis part. A. General Description In our work at Stanford, we are taking as a point of departure the Argonne work on automatic film data processing. That work was the collaborative effort of several people in the development of a film digitizer, programs for rurning the film digitizer on line to a large computer (CDC 3600) and the analysis programs for handling the digitized film data [Reference 21. Although the Argonne work was done explicitly for film data, the programs maintained sufficient generality that they can be used for any graphic data from spark chambers no matter how the data is pre- sented to the programs. Figure 4 shows the structure of these pro- grams. The scanning and measuring program AROMA prepares,for the AIRWICK programs,the digitized information from the photographs. The AIRKICK programs identify the corresponding sprks in the two stereo views, calculate their positions in three space, and then link these sprks into tracks. Figure 5 shows a stereo pair of photographs of an event taking place in a spark chamber and Figure 6 shows the results of two stages of processing - after AROMA and after AIRWICK. One of the most pleasing aspects of this work is that we can give a formal description of each step. Formally, the problem is presented as (a) generation of a graph,the vertices of which are given by the three space coordinates of the sparks, and (b) the selection Of the 6 proper tree (or forest of trees for multiple events) to represent the event. The processes physically described as (1) scanning, measur- ing, and image transformation, (2) pairing in the stereo views, and (3) linking into tracks have their counterparts in the formal des- cription. The first -wo of chebe ~rcees~s.. i-k*, SC%.&%, etc., and pairing, etc., are concerned with t:ie generation of the graph and the third, i.e., linking, etc., is concerned with the tree sel- ection. Figure 7 shows the relationship between the physical pro- cesses and the formal description. The essential elements of the sys- tem are described in brief detail below. Let me first give some data on the rates of manual and then of our automatic system. Using the manual systems and skilled human operators, one can expect to scan and measure at the rate of about 20 to 30 events per hour. We are able to scan and measure at about an order of ma2nitud.e greater speed. We can process a stereo view of ccmplex nature in about 20 seconds and simple ones in lo-12 seconds. B. Scanning, Measuring, and Image Transformation. The C'HLOE film digitizer is described elsewhere [References 2, 101 and will be described only briefly here. CHLOE is a hardware system for digitizing data recorded on transparent 35 mm. film. The hard- ware consists of (1) a controlling computer, (2) an optical scanner (CRT) operating under the control of the computer, and (3) a data link to a larger computer. A spot from the Cathode Ray Tube is pro- jected on the film and the transmitted light is measured by a photo- multiplier. From this measurement a decision is made concerning the density of any rectangular portion of a bo96xk-096 raster measuring 1.25 inches on a side. In practice,onescans only a small portion of this raster. -- -- - 9 The contextual parameters are order and interference and these para- meters enter D(i,j) through F2 and F 3 respectively. (4) F2(W) = 1 - Ii - JI p where P+A is the &ximum d2ens'LUn of '-hc blork- (5) F3(bj) = (l s(;)-l)(l T(2-1 ) where R is the maximum dimension of the block, S(S) is the number of geometrically possible pairings of spark i , and T(j) is the number of gecxnetrically possible pairings of spark j . The F's are all normalized to yield values in the internal (0,l) . The weights w1 7 w2 t and W 3 are experiment dependent. Large values of D(i,j) mean high probability of pairing. On the other hand, low values of A(i,j) mean high probability of pairing - a low value of A(i,j) means that there is low probabtlity that the ith spark of view one could be paired with any spark other than the jth spark of view two, or that the jth spark of view two could be paired with any spark other than the ith spark of view one. By repeated application of D(i,j) and A(i,j) successful pair- ings are determined. With each determination)the relationship matrix is reduced by one row and one column until all ambiguities are re- solved. It is possible that all ambiguities are not resolved and that some are left unresolved until the linking operation. The PAIRING ALGORITRM then passes on to the LINKING ALGORITHM a relationship matrix with a#Jmost all ambiguities removed. 10 D. Linking Into Tracks. The output of the PAIR program is a set of three space coordinates that form a graph. The LINK program selects the proper tree (or forest in the case of multiple events) to represent the particle tracks. The detailed tree &elect&W aJ,.gcr5tllll~.isgivea in Refe.rence 2, Briefly, edges are establishea between the variol's vr;rt;*ces of t& graph on the basis of a number of criteria, such as Euclidean distance, number of vertices, direction of the edge, linear and helical extrapolation, and geometry of the spark chamber. Finally, the subgrapl selected is the minimal connector tree that contains IW circuits. The deter- mination of the minimal connector tree is accomplished by an algorithm due to Kruskal iReference XL]. The gra@ data are stored in memory in a multi-word list. Each list item contains all the necessary information about a given spark. Each list item contains seventeen words in all, including the three coordinates of the spark, its gap number, its chamber number, the local degree of the vertex, pointers to as many as seven connecting sparks, distance and pointer to the closest spark, and distance and pointer to the second closest spark. Raving arrived at a set of coordinates that represent the paths of particles participating in the event the data are now sent on to fitting programs and programs that extract the physics information from the data. Iv. CONSLUSIONS In order to develop more complete systems of the type depicted in Figure 1, it will be necessary to incorporate complex analysis L- - - 11 programs of the type described in Section III above. This is by no means all of the story. The kinematic analysis programs and the hypothesis testing programs that follow are also very complex. How- ever, progress is being made and physics data are being analyzed at a very rapid rate. Moreover, in&penlent:iy VX% 2 proceeding on the control aspects. One developnt which I feel is not getting its full share of attention in these problems is in the area of displays, control consoles, and the system control languages. That is, the command posts are not being developed as thoroughly as they should be. The splendid use of graphics that we have seen in other areas could be put to excellent use in these large analysis and control systems. -- - CAPI'IONS FOR FIGURE3 .- 14 Figure 1. General Schematic of Integrated Data Analysis and Control System for Complex Experiments. Figure 2. Closed Loop Analysis 2nd Control Qys:em being Utilized in Low Energy Nuclear Physics Experiments. Figure 3. Schematic Flow Chart of First Program Run on the System Depicted by Figure 2. Figure 4. Schematic of the Flow of Data in the Argonne System. Figure 5. Photograph of Spark Chamber Event in Two Views. Figure 6. Output of AROMA and AIRWICK for Event Shown in Figure 5. Figure 7. Schematic of the Relationship of Physical Processes to Formal Description. Figure 8. Line Segments and Cells Generated by CELL CONSTRUCTION ALGORITHM. Figure 9. Schematic of Top of Spark Chamber with Five Sparks and Corresponding Relationship Matrix. __- - BOX 7 COMPLEX EXPEP!MENTAL SYSTEMS JUCH AS ’ ACCELERb.1 W’S ANC REwZ’Tr-3s BOX 6 DATA ACQUISITION DEVICES DATA REDUCTION AND ANALYSIS L BOX 4 CONTROL SYSTEM FOR EXPERIMENTAL EQUIPMENT J BOX 5 DECISION BOX AND EXPERIMENT PROGRAMMER f yYybA Fig. 1 NUCLEAR PROCESS FEEDBACK BY ION BEAM, f ( E,L, DIGITAL I I A-D IF REQUIRED I BUFFER HOLD I TRANSMIT ALL INFO. I I I OTHERS I I I I I ---- --- ---- -- I INFO. 2’2 ’ I- CoM. 1 I 1 I 1 I I L I I I I ______ ------I I I I I I BOXES 3 6 5 I l --- _------ Fig. 2 Fig. & Measurements from the Sam.e Two Photographs after Processing by AROMA “U :::z =:e ‘%:: “:I= z::::: 2::.::: : ,(.,.,, . ,,,.,. ;... .;. ,,.. .;i” :\ I * \: -:d I ,. . ..l “::= 2::::: “!:.:” ::::s ::::::: ‘::::::: ~ ,,.,. .,,...,..... i’ .,.._........_... I: : : I : : : : i : ; : : : : : : : : : : : : ., : : ,.,i :. i .I I.1 .,.%.” ““I,” ‘“:“: “.l” ..“I”. .y,. “I::;: .*:;:;~ :,,,,,, ,,,,,,..,(,,,....,,,(. ((,.,.,,,..,......,.,,,,...,,........................ ‘“..; -./ \: . : . ., .._..... I .!. : : : : : ; : : : : : : : : : : : : *; : : ; : : ; : ,..,.: :.. The Event after Processing by AIRWICK, Showing the Reconstructed Tracks. Random Sparks Not On the Tracks Have Now Been Eliminated. Fig. 6 PHYSICAL PROCESS SCANNING, MEASURING, AND IMAGE TRANSFORMATION FORMAL DESCRIPTION GRAPH GE#ERA;r/ON -------- ------- 1 EXECUTION OF THE SCANNING ALGORITHM I 1’1 GENERATION OF RELATIONSd’P MATRIX PAIRING SPARK IMAGES IN STEREO VIEWS GIVING “AIRED SPARK IMAGES. REDUCTION OF AMBIGUITIES BY ITERATIVE APPLICATION OF A PAIR DECISION FUNCTION AND ITS CO-MATRIX. I ---m-m-- --- ----_ - TEE SEL EC JION r--- ----- ----ey1 LINKING SPARKS INTO TRACKS I Z?- 3 I SELECTION OF MINIMAL CONNECTOR TREE OF A CIRCUIT FREE GRAPH I , --l L-m------------ Fig. 7 337-5 -A