Academia.eduAcademia.edu

Figure 93 – uploaded by spyros G tzafestas

See full PDFdownloadDownload figure
L4 1A OUULLULUDS | The structural element such as we consider here is found as a girder, a ship hull or a spar structure. Suppose an original design plan was to place a horizontal mem- ber to a longitudinal one as shown in Fig. 16.5. If we connect these two members this way, it is expected that we come up with such problems of deteriorated weld quality and stress concentration. As the weldlines meet at one point marked in the figure as dangerous point, the thermal history there becomes very complicated and thus it would possibly deteriorate the weld quality there and what makes the matter worse this point also serves as start and stop points for weld arcs. These are the problems on the strength side but on the loading side there is a problem of stress concentration.

Figure 16 L4 1A OUULLULUDS | The structural element such as we consider here is found as a girder, a ship hull or a spar structure. Suppose an original design plan was to place a horizontal mem- ber to a longitudinal one as shown in Fig. 16.5. If we connect these two members this way, it is expected that we come up with such problems of deteriorated weld quality and stress concentration. As the weldlines meet at one point marked in the figure as dangerous point, the thermal history there becomes very complicated and thus it would possibly deteriorate the weld quality there and what makes the matter worse this point also serves as start and stop points for weld arcs. These are the problems on the strength side but on the loading side there is a problem of stress concentration.

Related Figures (139)

Fig. 1.1. Snapshot of a sample rule-based expert system
Fig. 1.1. Snapshot of a sample rule-based expert system
Fig. 1.2. Example of a frame based system in the transportation domain
Fig. 1.2. Example of a frame based system in the transportation domain
Table 1.1. Correspondence between AI concepts and simulation concepts elling and studying complex systems), they differ in the way they function. An ES gives a prescription; a simulation model provides a prediction. That is, given a goal, an ES suggests a course of action, whereas a simulation model predicts the consequences of a selected course of action under a certain simulation scenario. A convenient way for merging and unifying the ES and simulation models is proposed by Reddy in [45] through the use of concepts like “objects” and “classes” first introduced in Simula 67., and more recently, in SMALLTALK and LOOPS in refined forms (using new terminology such as messages, inheritance, demons, and worlds/contexts). Table 1.1 illustrates the correspondence between AI concepts and simulation concepts. Almost all of the concepts in AI may be used for some aspect of simulation;
Table 1.1. Correspondence between AI concepts and simulation concepts elling and studying complex systems), they differ in the way they function. An ES gives a prescription; a simulation model provides a prediction. That is, given a goal, an ES suggests a course of action, whereas a simulation model predicts the consequences of a selected course of action under a certain simulation scenario. A convenient way for merging and unifying the ES and simulation models is proposed by Reddy in [45] through the use of concepts like “objects” and “classes” first introduced in Simula 67., and more recently, in SMALLTALK and LOOPS in refined forms (using new terminology such as messages, inheritance, demons, and worlds/contexts). Table 1.1 illustrates the correspondence between AI concepts and simulation concepts. Almost all of the concepts in AI may be used for some aspect of simulation;
Table 1.2. Numerical representation (max-min = 100%) of state classification OR relative speed is tiny OR relative speed is small OR relative speed is big OR relative speed is huge THEN relative speed classification IF quadrant is classified IF error is classified IF speed/error is classified THEN state classification completed
Table 1.2. Numerical representation (max-min = 100%) of state classification OR relative speed is tiny OR relative speed is small OR relative speed is big OR relative speed is huge THEN relative speed classification IF quadrant is classified IF error is classified IF speed/error is classified THEN state classification completed
Table 2.1. An information table example b « B has a specified value associated to each equivalence class X. This is callec a description of the class X, denoted by
Table 2.1. An information table example b « B has a specified value associated to each equivalence class X. This is callec a description of the class X, denoted by
Table 2.2. Budget Form for Problem 4 (Inflation Rate 8%) Note: The problem is to fill in the blank positions for next year’s budget.
Table 2.2. Budget Form for Problem 4 (Inflation Rate 8%) Note: The problem is to fill in the blank positions for next year’s budget.
Table 2.3. Actual spending in the budget items for present year
Table 2.3. Actual spending in the budget items for present year
Fig. 3.1. Fault identification
Fig. 3.1. Fault identification
Fig. 3.2. Test tree display
Fig. 3.2. Test tree display
Fig. 3.6. Test tree for sequential circuit in Fig. 3.5
Fig. 3.6. Test tree for sequential circuit in Fig. 3.5
Table 3.1. Test tree algorithm performance in the purely sequential case 3.3.2.2 Purely Parallel Connections
Table 3.1. Test tree algorithm performance in the purely sequential case 3.3.2.2 Purely Parallel Connections
Table 3.2. Failure probabilities of single components and their associated costs of testing In general, however, the best test will always be chosen as a non-terminal test. In particular, when all the failure probabilities are equal and the costs of the tests are equal, the main equation 3.2 will reduce to the familiar bisection method, producing a balanced test tree. So we may claim that a variation of a “group testing” methodology is implemented via utilization of the main equation 3.2: a test will
Table 3.2. Failure probabilities of single components and their associated costs of testing In general, however, the best test will always be chosen as a non-terminal test. In particular, when all the failure probabilities are equal and the costs of the tests are equal, the main equation 3.2 will reduce to the familiar bisection method, producing a balanced test tree. So we may claim that a variation of a “group testing” methodology is implemented via utilization of the main equation 3.2: a test will
3.4 Reification of Expert Systems Among the important aspects of the knowledge acquisition problem are knowl- edge base inconsistency and redundancy. At all places of knowledge acquisition processes, the developers must ensure the consistency and nonredundancy between the rules. A knowledge base is inconsistent when there is a set of data (inputs) which, when supplied it contains rules which can be removed without conclusions. A knowledge base is redundant when it contains rules which can be removed without changing the set of possible conclusions. a A “ ais 4 ii
3.4 Reification of Expert Systems Among the important aspects of the knowledge acquisition problem are knowl- edge base inconsistency and redundancy. At all places of knowledge acquisition processes, the developers must ensure the consistency and nonredundancy between the rules. A knowledge base is inconsistent when there is a set of data (inputs) which, when supplied it contains rules which can be removed without conclusions. A knowledge base is redundant when it contains rules which can be removed without changing the set of possible conclusions. a A “ ais 4 ii
Table 4.1. Expert systems dealing with digital systems [2]
Table 4.1. Expert systems dealing with digital systems [2]
upgrade of the hierarchical logic design support systems to the expert designer’s level.
upgrade of the hierarchical logic design support systems to the expert designer’s level.
which is a knowledge-based tool. TALIB utilizes rules organized around the con- cepts of subtasks and demons. The demons select subtasks and other activities.
which is a knowledge-based tool. TALIB utilizes rules organized around the con- cepts of subtasks and demons. The demons select subtasks and other activities.
Table 4.2. VHSIC chips
Table 4.2. VHSIC chips
Fig. 4.4. Classification of VHSIC chips
Fig. 4.4. Classification of VHSIC chips
Fig. 4.5. Classification of memory chips
Fig. 4.5. Classification of memory chips
Fig. 4.6. Classification of processor chips
Fig. 4.6. Classification of processor chips
Fig. 4.7. Program flow control
Fig. 4.7. Program flow control
Fig. 4.8. RAM prototype slot characterization
Fig. 4.8. RAM prototype slot characterization
en ee To develop new hardware in the VHDL design environment, first a VHDL ource code is developed. The VHDL source code specifies the performance re- juirements of the target hardware system. It can be at one or more levels of a ystem design hierarchy such as system level, component level, logic level, gate evel, or below. The VHDL analyzer translates the source code into an intermedi- te form known as the abstract syntax tree (AST) which is incorporated into the lesign library. Based on user instructions, the profiler generates an AST represen- ation which reflects the correct configuration (i.e. entities with the correct revisions elected). The output of the profiler is incorporated into the design library and pre- ented to the hardware synthesizer tool. The hardware synthesizer then produces a ircuit design.
en ee To develop new hardware in the VHDL design environment, first a VHDL ource code is developed. The VHDL source code specifies the performance re- juirements of the target hardware system. It can be at one or more levels of a ystem design hierarchy such as system level, component level, logic level, gate evel, or below. The VHDL analyzer translates the source code into an intermedi- te form known as the abstract syntax tree (AST) which is incorporated into the lesign library. Based on user instructions, the profiler generates an AST represen- ation which reflects the correct configuration (i.e. entities with the correct revisions elected). The output of the profiler is incorporated into the design library and pre- ented to the hardware synthesizer tool. The hardware synthesizer then produces a ircuit design.
Fig. 4.15. Schematic diagram for an ALU
Fig. 4.15. Schematic diagram for an ALU
Table 5.1. Characteristics of LANDSAT TM images
Table 5.1. Characteristics of LANDSAT TM images
Fig. 5.2a-g. Examples of LANDSAT TM Images. a Band 1; b Band 2; ¢ Band 3; d Band 4; e Band 5; f Band 6; g Band 7
Fig. 5.2a-g. Examples of LANDSAT TM Images. a Band 1; b Band 2; ¢ Band 3; d Band 4; e Band 5; f Band 6; g Band 7
identified as a thin object which is a protrusion of S. In this system, rivers (C;’s are separated from sea areas (S* parts) according to this principle. In this case, deciding the value k is a key point, because an optimal valu depends on the scale of rivers. A relatively large value is required for wide river: However, narrow rivers could sometimes be merged if the same value is used. T solve this problem, the system examines whether there is any difference betwee 5 — S* and S — S*” by increasing continuously the value k, and if theré is n difference, this kind of regions will not be processed further. The separation proces terminates when this condition becomes valid for every connected component water regions. TT eee s- 1 . - /? ee we
identified as a thin object which is a protrusion of S. In this system, rivers (C;’s are separated from sea areas (S* parts) according to this principle. In this case, deciding the value k is a key point, because an optimal valu depends on the scale of rivers. A relatively large value is required for wide river: However, narrow rivers could sometimes be merged if the same value is used. T solve this problem, the system examines whether there is any difference betwee 5 — S* and S — S*” by increasing continuously the value k, and if theré is n difference, this kind of regions will not be processed further. The separation proces terminates when this condition becomes valid for every connected component water regions. TT eee s- 1 . - /? ee we
where J() means brightness of a pixel, and Stdv() means standard deviation. The first equation is the same as the one used in the narrow river extraction (see 5.4.2.3). The second equation is newly augmented here based on Characteris- tic 2 in 5.4.3.1, in order to prevent the system from extracting noise elements. It means “when the intensity variance of the center of a template is high, (i.e., when it does not have uniform intensity), the center should not be judged to be a part of road, even if it has high edgeness.” In this method, the edge strength is a maximum E, which satisfies the first and the second equations, and the edge direction is the direction of a template which gives the maximum Ey. The oneration is. in practice. time consumins. This is because when the size of GUCCUOT! OL da LCP late WINCH SIVes WIC WIG AUT Eg. The operation is, in practice, time consuming. This is because when the size of edge templates is 3 x n, the system has to match 4 x (nm — 1) templates with the image in order to calculate edgeness of one pixel. We, therefore, adopt hierarchical calculation in order to reduce computation time. In the calculation, the primary edgeness is firstly calculated on sparse masks with the size of 3 x 3. If the edge strength is large enough, the exact edgeness is calculated on the masks with the size of 3 x n. Otherwise, the system judges that there is no edge primitives.
where J() means brightness of a pixel, and Stdv() means standard deviation. The first equation is the same as the one used in the narrow river extraction (see 5.4.2.3). The second equation is newly augmented here based on Characteris- tic 2 in 5.4.3.1, in order to prevent the system from extracting noise elements. It means “when the intensity variance of the center of a template is high, (i.e., when it does not have uniform intensity), the center should not be judged to be a part of road, even if it has high edgeness.” In this method, the edge strength is a maximum E, which satisfies the first and the second equations, and the edge direction is the direction of a template which gives the maximum Ey. The oneration is. in practice. time consumins. This is because when the size of GUCCUOT! OL da LCP late WINCH SIVes WIC WIG AUT Eg. The operation is, in practice, time consuming. This is because when the size of edge templates is 3 x n, the system has to match 4 x (nm — 1) templates with the image in order to calculate edgeness of one pixel. We, therefore, adopt hierarchical calculation in order to reduce computation time. In the calculation, the primary edgeness is firstly calculated on sparse masks with the size of 3 x 3. If the edge strength is large enough, the exact edgeness is calculated on the masks with the size of 3 x n. Otherwise, the system judges that there is no edge primitives.
Fig. 5.10a-c. Graph modification rules. a Rule for deltas; b Rule for shoals; ¢ bridge finding rule 5 Expert System of Image Processing and its Application to LANDSAT Image Analysis
Fig. 5.10a-c. Graph modification rules. a Rule for deltas; b Rule for shoals; ¢ bridge finding rule 5 Expert System of Image Processing and its Application to LANDSAT Image Analysis
Figure 5.13 shows an example of the road extraction process. (a) is an original image (Band 4), (b) is the result of the edge clustering and the line segment descrip- ‘ion, and (c) shows roads certificated by the collinearity and the perpendicularity criteria. (d) shows another example, which is acquired by applying this method io Fig. 5.2 images. (Dashed lines show the connections among road segments. Triangle symbols show assumed bridges after the bridge finding rule is applied.) These results show good performance of this road extraction process. A few narrow river courses are misidentified, because lower reaches of a river are not extracted in the scene, and therefore, narrow upper reaches cannot be precisely extracted. The problem left is that all of the roads in town areas are not extracted by this method. It is owing to the fact that the spectral distribution of town areas and that of roads are almost the same, and it is quite difficult even for a human interpreter to discriminate roads from other town areas. Bionre & 14 echaue 9a racult af 49 nreliminary ayvnerimant af mainntain chana
Figure 5.13 shows an example of the road extraction process. (a) is an original image (Band 4), (b) is the result of the edge clustering and the line segment descrip- ‘ion, and (c) shows roads certificated by the collinearity and the perpendicularity criteria. (d) shows another example, which is acquired by applying this method io Fig. 5.2 images. (Dashed lines show the connections among road segments. Triangle symbols show assumed bridges after the bridge finding rule is applied.) These results show good performance of this road extraction process. A few narrow river courses are misidentified, because lower reaches of a river are not extracted in the scene, and therefore, narrow upper reaches cannot be precisely extracted. The problem left is that all of the roads in town areas are not extracted by this method. It is owing to the fact that the spectral distribution of town areas and that of roads are almost the same, and it is quite difficult even for a human interpreter to discriminate roads from other town areas. Bionre & 14 echaue 9a racult af 49 nreliminary ayvnerimant af mainntain chana
Fig. 5.13a-d. Result of Road Extraction. a Original image; b result of edge clustering; ¢ final result; d another experimental result
Fig. 5.13a-d. Result of Road Extraction. a Original image; b result of edge clustering; ¢ final result; d another experimental result
Fig. 5.14a,b. Preliminary experiment of mountain shape analysis. a Extracted mountain areas; b result of surface recovery
Fig. 5.14a,b. Preliminary experiment of mountain shape analysis. a Extracted mountain areas; b result of surface recovery
When the flows are fixed, and the cost of the links linearly depends on the capacities, the problem becomes one of determining the capacity vector, C, which minimizes the cost of the topology and satisfies the constraint of delay (T < Twax). It, then, becomes an isolated problem of capacity assignment for linear link costs, d;(C;). This makes an optimal solution possible, by using the method of Lagrange multipliers. The optimal capacities, C7, and the optimal cost, D*, are given by the following relations [2]:
When the flows are fixed, and the cost of the links linearly depends on the capacities, the problem becomes one of determining the capacity vector, C, which minimizes the cost of the topology and satisfies the constraint of delay (T < Twax). It, then, becomes an isolated problem of capacity assignment for linear link costs, d;(C;). This makes an optimal solution possible, by using the method of Lagrange multipliers. The optimal capacities, C7, and the optimal cost, D*, are given by the following relations [2]:
Fig. 6.6. Diagram of an expert system
Fig. 6.6. Diagram of an expert system
4 Samuel Pierre and Evelyne Hausen-Tropper
4 Samuel Pierre and Evelyne Hausen-Tropper
6.5.2.1 Basic Definitions The rules may be classified into two categories: deterministic and nondeterministic rules. Deterministic rules serve to describe absolute truth contexts, and are con- sequently accompanied by likelihood factors equal to one. On the other hand, a rule is nondeterministic when it refers to an uncertainty situation, expressed by a likelihood factor less than one. Obviously, the likelihood factors are nonnegative real numbers not greater than one.
6.5.2.1 Basic Definitions The rules may be classified into two categories: deterministic and nondeterministic rules. Deterministic rules serve to describe absolute truth contexts, and are con- sequently accompanied by likelihood factors equal to one. On the other hand, a rule is nondeterministic when it refers to an uncertainty situation, expressed by a likelihood factor less than one. Obviously, the likelihood factors are nonnegative real numbers not greater than one.
Table 7.1. Summary of surveyed systems
Table 7.1. Summary of surveyed systems
Table 7.1. Summary of surveyed systems (continued)
Table 7.1. Summary of surveyed systems (continued)
A: Search Strategy, B: Placement Table 8.1. Examples of knowledge
A: Search Strategy, B: Placement Table 8.1. Examples of knowledge
Fig. 8.3a,b. AND/OR graph representing layout modification
Fig. 8.3a,b. AND/OR graph representing layout modification
Yasuhiro Kobayashi, Kenichi Yoshida and Toshiaki Yoshinaga
Yasuhiro Kobayashi, Kenichi Yoshida and Toshiaki Yoshinaga
Fig. 8.5. Knowledge base configuration
Fig. 8.5. Knowledge base configuration
Fig. 8.6. Example of data on work area
Fig. 8.6. Example of data on work area
with all the modified objects. As all conditions can be verified, the hypothesis is realized as below:
with all the modified objects. As all conditions can be verified, the hypothesis is realized as below:
Fig. 8.8. Modified layout of components and pipes
Fig. 8.8. Modified layout of components and pipes
control in the compressor assembly factory. The quality control engineers are able to obtain the information from EXACT to detect the cause of a production fault in the compressor inspection process. This may result in an increase in the reliability and productivity of the compressor manufacturing process. Since the personnel re- source in any organization is not infinite, the use of expert systems has a significant impact cn the personnel management [13].
control in the compressor assembly factory. The quality control engineers are able to obtain the information from EXACT to detect the cause of a production fault in the compressor inspection process. This may result in an increase in the reliability and productivity of the compressor manufacturing process. Since the personnel re- source in any organization is not infinite, the use of expert systems has a significant impact cn the personnel management [13].
Table 9.1. Expert Systems for Troubleshooting
Table 9.1. Expert Systems for Troubleshooting
Fig. 9.3. Combining two certainty factors [Modified from P. Harmon, 1985]
Fig. 9.3. Combining two certainty factors [Modified from P. Harmon, 1985]
SES! LAER SO SRT |) TT mR In the production fault database, each record has 22 fields consisting of the type of compressor, a beginning production number, an ending production number, and 19 compressor components. The cumulative historical database is divided into three fields which consist of the type of compressor, the cause of trouble, and its number of occurrences. Based on the current type of trouble, Locking, Noise, Leakage, and Poor performance, a percentage frequency of each cause is computed.
SES! LAER SO SRT |) TT mR In the production fault database, each record has 22 fields consisting of the type of compressor, a beginning production number, an ending production number, and 19 compressor components. The cumulative historical database is divided into three fields which consist of the type of compressor, the cause of trouble, and its number of occurrences. Based on the current type of trouble, Locking, Noise, Leakage, and Poor performance, a percentage frequency of each cause is computed.
Fig. 9.5. An example of pictorial screen with its menu
Fig. 9.5. An example of pictorial screen with its menu
Fig. 9.6. Menu driven interface in EXACT
Fig. 9.6. Menu driven interface in EXACT
9.5 System Performance Validation Two factors were used to compare the results provided by EXACT versus those given by the human expert. One factor is the percentage accuracy with which each evaluator provided the true cause as his estimation of the most probable cause of the compressor trouble. This most probable cause is represented as the “Ist” rank in the diagnosis of the 50 sample cases. The “2nd” rank and the “3rd” rank for the probable cause means that the evaluator estimated it as the second highest cause and the third highest probable cause respectively. If the evaluator does not provide the true cause as the first probable cause, then the inspector will have to check the next probable cause provided by the evaluator. The inspector will have to continue this process until he/she finds the true cause of the trouble.
9.5 System Performance Validation Two factors were used to compare the results provided by EXACT versus those given by the human expert. One factor is the percentage accuracy with which each evaluator provided the true cause as his estimation of the most probable cause of the compressor trouble. This most probable cause is represented as the “Ist” rank in the diagnosis of the 50 sample cases. The “2nd” rank and the “3rd” rank for the probable cause means that the evaluator estimated it as the second highest cause and the third highest probable cause respectively. If the evaluator does not provide the true cause as the first probable cause, then the inspector will have to check the next probable cause provided by the evaluator. The inspector will have to continue this process until he/she finds the true cause of the trouble.
Table 9.2. An Example of the List of the Problem Causes
Table 9.2. An Example of the List of the Problem Causes
Table 9.3. Results of validation test of the human expert. The human expert found the true cause of the trouble in the compressor faster than EXACT for 6 sample cases, while EXACT found the true cause faster than the human expert for 8 sample cases. In average, the time consumed by EXACT are shorter than that of the human expert. And for the other 36 cases both the human expert and EXACT yielded equally fast results. In order to compare the performances of EXACT and the human expert obtained from the test, the nonparametric sign test based on the binomial distribution was used [3]. The following hypothesis was made for the sign test which was a two-tailed test,
Table 9.3. Results of validation test of the human expert. The human expert found the true cause of the trouble in the compressor faster than EXACT for 6 sample cases, while EXACT found the true cause faster than the human expert for 8 sample cases. In average, the time consumed by EXACT are shorter than that of the human expert. And for the other 36 cases both the human expert and EXACT yielded equally fast results. In order to compare the performances of EXACT and the human expert obtained from the test, the nonparametric sign test based on the binomial distribution was used [3]. The following hypothesis was made for the sign test which was a two-tailed test,
Fig. 10.1a-f. Data released by the monitoring system Jean-Marc David, Jean-Paul Krivine, Jean-Pierre Tiarri and Benoit Ricard
Fig. 10.1a-f. Data released by the monitoring system Jean-Marc David, Jean-Paul Krivine, Jean-Pierre Tiarri and Benoit Ricard
From this example, we can see that it is not indifferent to choose one or the other way; the last query - if possible - will deliver the most precise answer to our question, but will disrupt a lot the exploitation. On the other hand, a request to the historical data base is quite cheap, but is obviously less informative. Among the necessary knowledge this task needs about each piece of information, the different ways to retrieve or infer information and the strategy to follow are critical ones. Thus, DIVA is built from three main tasks. Each of these tasks can then be further decomposed into sub-tasks; this decomposition is pursued till tasks can no longer be decomposed or decomposition is of no relevance (Fig. 10.4). The strategy of a task is the way this task controls its sub-tasks and the application of
From this example, we can see that it is not indifferent to choose one or the other way; the last query - if possible - will deliver the most precise answer to our question, but will disrupt a lot the exploitation. On the other hand, a request to the historical data base is quite cheap, but is obviously less informative. Among the necessary knowledge this task needs about each piece of information, the different ways to retrieve or infer information and the strategy to follow are critical ones. Thus, DIVA is built from three main tasks. Each of these tasks can then be further decomposed into sub-tasks; this decomposition is pursued till tasks can no longer be decomposed or decomposition is of no relevance (Fig. 10.4). The strategy of a task is the way this task controls its sub-tasks and the application of
Fig. 11.3. Screen display of conversational interface for sludge symptom implementation of intelligent control systems, suggested by [7]. Second, the finite automaton formalism is particularly natural to implemen- tation in PROLOG. A simple meta-interpreter combined with a database of triplets delta(state,input,new-state) constitute a straightforward implementation. This methodology has been used in practice to implement an intelligent control system for alkaline etch in a printed wiring board factory. The system serves simul- taneously two purposes: first it dispenses on-line expert advice on how to control and diagnose chemical etching; and secondly, it constitutes a repository of quali- tative knowledge, and thus serves to codify the folklore of alkaline etch control.
Fig. 11.3. Screen display of conversational interface for sludge symptom implementation of intelligent control systems, suggested by [7]. Second, the finite automaton formalism is particularly natural to implemen- tation in PROLOG. A simple meta-interpreter combined with a database of triplets delta(state,input,new-state) constitute a straightforward implementation. This methodology has been used in practice to implement an intelligent control system for alkaline etch in a printed wiring board factory. The system serves simul- taneously two purposes: first it dispenses on-line expert advice on how to control and diagnose chemical etching; and secondly, it constitutes a repository of quali- tative knowledge, and thus serves to codify the folklore of alkaline etch control.
12 Knowledge-Based Adaptive Identification for Process Control and Modelling 2 The objective of intelligent approaching is to move I, into I], through [,. The parameters of IT, are initially selected by the operator using the expert system, during consultation. The procedures of the intelligent approaching are given as follows:
12 Knowledge-Based Adaptive Identification for Process Control and Modelling 2 The objective of intelligent approaching is to move I, into I], through [,. The parameters of IT, are initially selected by the operator using the expert system, during consultation. The procedures of the intelligent approaching are given as follows:
Fig. 14.2. Model of Expert Assisted Robot Skill Acquisition
Fig. 14.2. Model of Expert Assisted Robot Skill Acquisition
Expert skills are represented by a set of production rules, termed the “expert rules”. The condition and action parts of the expert rules are described in terms of a set of primitive features selected by the expert to define situation and action spaces, respectively. Table 14.1 shows a set of primitive features selected for the expert situation and action spaces. A rule is also associated with a certainty factor provided
Expert skills are represented by a set of production rules, termed the “expert rules”. The condition and action parts of the expert rules are described in terms of a set of primitive features selected by the expert to define situation and action spaces, respectively. Table 14.1 shows a set of primitive features selected for the expert situation and action spaces. A rule is also associated with a certainty factor provided
Table 14.1. A set of primitive features selected for the expert situation space and the action space
Table 14.1. A set of primitive features selected for the expert situation space and the action space
Table 14.3. A set of primitive features selected for the robot action space the actual robot motion to be applied. In practice, we select a small size of window, e.g. one or a few intervals, in defining robot action. If the size of the window is reduced to one, no distinction exists between a robot action and a robot motion. A robot motion is defined by a triple {V,, Vy, W.}, where V, and V, represent the linear velocities of an end effector along the x and y axes and W, represents the angular velocity about the z axis.
Table 14.3. A set of primitive features selected for the robot action space the actual robot motion to be applied. In practice, we select a small size of window, e.g. one or a few intervals, in defining robot action. If the size of the window is reduced to one, no distinction exists between a robot action and a robot motion. A robot motion is defined by a triple {V,, Vy, W.}, where V, and V, represent the linear velocities of an end effector along the x and y axes and W, represents the angular velocity about the z axis.
The developed robot skill learning is incremental, i.e. hypotheses are generated teratively every learning cycle. Incremental learning not only provides adaptability 0 environmental change but also accelerates the learning process by the direct ecycling of whatever was learned. However, this requires the active management f the hypotheses and rules generated, including the monitoring of their credibility ased on the new evidence from subsequent experiences. Figure 14.5 outlines the verall structure of robot fine motion skill learning.
The developed robot skill learning is incremental, i.e. hypotheses are generated teratively every learning cycle. Incremental learning not only provides adaptability 0 environmental change but also accelerates the learning process by the direct ecycling of whatever was learned. However, this requires the active management f the hypotheses and rules generated, including the monitoring of their credibility ased on the new evidence from subsequent experiences. Figure 14.5 outlines the verall structure of robot fine motion skill learning.
Fig. 14.7. Hypothesis generation based on the generation tree method. (h, < 0.1 is used as the criterion of Success-Node, and the Search-Depth=3. The fractional numbers at each node represent the value of he)
Fig. 14.7. Hypothesis generation based on the generation tree method. (h, < 0.1 is used as the criterion of Success-Node, and the Search-Depth=3. The fractional numbers at each node represent the value of he)
Table 14.5. Peg and hole size if Pe(t;) > Pe(tig1) and P.(t;) > Pe(ti+2) as Negative, if Pe (ti) F Pe (titi) and Pe(ti) 7 Pe(ti+2) else Undecided Table 14.4. Primitive features and feature values selected for robot action space
Table 14.5. Peg and hole size if Pe(t;) > Pe(tig1) and P.(t;) > Pe(ti+2) as Negative, if Pe (ti) F Pe (titi) and Pe(ti) 7 Pe(ti+2) else Undecided Table 14.4. Primitive features and feature values selected for robot action space
Fig. 14.10. The comparison of peg insertion skill (a) at the initial trial and (b) at the final trial (initial peg position A)
Fig. 14.10. The comparison of peg insertion skill (a) at the initial trial and (b) at the final trial (initial peg position A)
Fig. 15.1. Components of a mechanical assembly planning system By combining all these techniques the system should be able to generate its as- sembly plan, to formulate the manipulator movement, and finally to accomplish the assembly task.
Fig. 15.1. Components of a mechanical assembly planning system By combining all these techniques the system should be able to generate its as- sembly plan, to formulate the manipulator movement, and finally to accomplish the assembly task.
We used Prolog because it can not only represent topology in a very concise form but it is also very convenient for attaching attributes. Figure 16.3 shows Prolog representation. The topological representation of a structure shown in Fig. 16.1 is represented as in Fig. 16.2. A face is defined by more than one loop and each loop is defined by the edges placed in the counterclockwise order. This order or the direction is taken so as to coincide with the rotational direction of a right-handed screw whose moving direction is from the inside out. Multiple face loops appear when there is a contact face or a hole.
We used Prolog because it can not only represent topology in a very concise form but it is also very convenient for attaching attributes. Figure 16.3 shows Prolog representation. The topological representation of a structure shown in Fig. 16.1 is represented as in Fig. 16.2. A face is defined by more than one loop and each loop is defined by the edges placed in the counterclockwise order. This order or the direction is taken so as to coincide with the rotational direction of a right-handed screw whose moving direction is from the inside out. Multiple face loops appear when there is a contact face or a hole.
Fig. 16.8. Decrease in the number of times of work positioning
Fig. 16.8. Decrease in the number of times of work positioning
What we should like to emphasize in this chapter is that the problem of shape is very important in considering manufacturability and the quality of a product, and that we have to make efforts to incorporate the factor of shape appropriately at every stage of production from design to manufacturing. If we could be successful, then it is expected that cooperative product development could be achieved with more ease. What we have described here are only preliminary attempts toward that goal.
What we should like to emphasize in this chapter is that the problem of shape is very important in considering manufacturability and the quality of a product, and that we have to make efforts to incorporate the factor of shape appropriately at every stage of production from design to manufacturing. If we could be successful, then it is expected that cooperative product development could be achieved with more ease. What we have described here are only preliminary attempts toward that goal.
Fig. 17.7. ako = a kind of : hierarchical link (conceptual hierarchy); is-a : link of instantiation
Fig. 17.7. ako = a kind of : hierarchical link (conceptual hierarchy); is-a : link of instantiation
Fig. 17.11. An example of different points of view on hierarchy
Fig. 17.11. An example of different points of view on hierarchy
Table 18.2. Maximum processing capacities of the FMSs
Table 18.2. Maximum processing capacities of the FMSs
Table 18.1. Specification of the lots to be processed
Table 18.1. Specification of the lots to be processed
Table 18.3. Specification of the final lot schedule
Table 18.3. Specification of the final lot schedule
Table 18.4. Final pattern of the production rates P(.), for each lot, for each FMS and each processing interval
Table 18.4. Final pattern of the production rates P(.), for each lot, for each FMS and each processing interval
Fig. 19.1. Typical layout of an unmanned robot cell
Fig. 19.1. Typical layout of an unmanned robot cell
Fig. 19.4. Control algorithm of one part at the machine
Fig. 19.4. Control algorithm of one part at the machine
Fig. 19.6. Control algorithm of process completed
Fig. 19.6. Control algorithm of process completed
An intelligent controller needs five system components: database, knowledge base, network interface, local machine interface, and supervisor. Figure 19.8 shows the system architecture. The database records information about the parts and the man- ufacturing system such as physical data (dimensions, shapes, weights, etc.) of the parts, process plan, cell configuration, etc. Currently, it is implemented in DBASE III. The knowledge base contains information of manufacturing strategy such as the cell control algorithm described in the previous sections, the role of a machine within the cell, the process constraints of a machine, etc. The PC-AT’s network interface handles the communication to and from the network [9]. This interface follows MAP standard and is implemented in the C programming language. The local machine interface links a PC-AT to the machine under its control and conveys process commands, machine status and emergency signals [15]. This local machine interface is also implemented in C. The supervisor, the last component of the sys- tem, coordinates the tasks of other system components. Both the supervisor and
An intelligent controller needs five system components: database, knowledge base, network interface, local machine interface, and supervisor. Figure 19.8 shows the system architecture. The database records information about the parts and the man- ufacturing system such as physical data (dimensions, shapes, weights, etc.) of the parts, process plan, cell configuration, etc. Currently, it is implemented in DBASE III. The knowledge base contains information of manufacturing strategy such as the cell control algorithm described in the previous sections, the role of a machine within the cell, the process constraints of a machine, etc. The PC-AT’s network interface handles the communication to and from the network [9]. This interface follows MAP standard and is implemented in the C programming language. The local machine interface links a PC-AT to the machine under its control and conveys process commands, machine status and emergency signals [15]. This local machine interface is also implemented in C. The supervisor, the last component of the sys- tem, coordinates the tasks of other system components. Both the supervisor and
Related topics:
Expert SystemsKnowledge-Based Expert SystemsIntelligent and Expert Systems
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved