JP2007193456A - Factor estimation system, factor estimation program, recording medium for recording factor estimation program, and factor estimation method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To optimize work sharing and work sequence of man and machine for man-machine cooperative diagnosing. <P>SOLUTION: A process management system 10 comprises an input item selecting section 64 which includes a predicted variation of certainty factor calculating section 64a for finding difference between a predicted certainty factor which is a certainty factor in a state that a characteristic quantity is set up to one of uninput conditions (target uninput condition) and the current certainty factor for each bad factor, and an input effect calculating section 64b where a user determines the sum of the predicted variation of certainty factors as the characteristic quantity corresponding to the target uninput condition, finds the input effect of the target uninput condition calculated by dividing the characteristic quantity with either smaller one of user cost for inputting cost and machine cost for calculating the characteristic quantity corresponding to the target uninput condition based on result data of investigation for every target uninput condition, and determines the uninput condition for maximizing the input effect. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a factor estimating apparatus, a factor estimating program, a recording medium storing a factor estimating program, and a factor estimating method for estimating a factor of a defect from a defect result generated in a diagnosis target system.

  In a factory production line, process improvement processing is required to improve the yield. As the process improvement process, first, a process that causes a defect of a manufactured product is specified, and adjustment or cleaning of the device is performed so as to remove the cause.

  However, in a manufacturing process composed of a plurality of processes, various factors such as a defect in a part of the manufacturing apparatus, a problem in setting the manufacturing apparatus, and a problem in the conveyance path can be considered as candidates for the cause of failure. For example, the circuit board surface mounting system process is divided into a printing process, a mounting process, and a reflow process. In the printing process, solder paste is applied on the substrate, and in the mounting process, components are installed on the substrate. In the final reflow process, the solder is melted by applying heat to bond the components. In such a surface mounting system, when a bridging failure occurs, there are many possible causes of bridging failure such as mask misalignment and lower mold contamination, but one or more of these are fundamental factors.

  When a phenomenon that causes a defect appears, not only does a symptom of a defect appear in a manufactured product, but it also has some influence on the operation history of the manufacturing apparatus and the inspection history of the inspection apparatus. The data regarding the symptoms of these defective products, and the data regarding the operation history of the manufacturing apparatus and the inspection history of the inspection apparatus become enormous, and it is difficult to analyze the occurrence of defects.

  Here, production managers (experts) who have abundant experience in production management know empirically the relationship between the effects of defective factors on defective products, manufacturing equipment, and inspection equipment, and how to interpret the effects. Therefore, it is possible to efficiently improve the process. However, inexperienced production managers must examine the factors one by one to identify the factors, and spend a lot of time improving the process.

  In addition, although experts have been performing failure diagnosis of devices that require advanced know-how and techniques and identifying failure factors for process improvement, there is a concern that there will be a shortage of skilled workers, which is said to be a 2007 problem. For this reason, there is a demand for making it possible for even beginners to estimate factors at the expert level.

  Therefore, there is a demand for a technique capable of realizing the estimation of the cause of abnormality with high accuracy and high efficiency, regardless of the skill level of the production manager in the production site. As such a technique, Patent Document 1 shown below proposes a method for obtaining information from an inspection machine provided in each process (printing process-mounting process-reflow process) and performing inference by that. .

  However, in the surface mounting system, not all information necessary for factor estimation can be acquired from the inspection machine. For example, the solder area taken from the front of the substrate can be collected by using image processing technology, but information such as the warpage of the substrate needs to be confirmed by a human from the side of the substrate. Here, it is conceivable to provide a configuration for detecting the warpage of the substrate, but it is not preferable to increase the cost of the inspection machine more than necessary. Therefore, it is preferable to input information from a human in addition to inspection result data from the inspection machine. That is, in order to perform factor estimation, it is necessary to perform cooperative inference between machines and people.

When factor estimation is performed by inputting information from a human, how to reduce the number of questions and perform reasoning accurately becomes a problem. For example, Patent Document 2 shown below proposes a method of giving an evaluation value based on the occurrence frequency in advance as a method of determining the question order.
Japanese Patent Laid-Open No. 6-196900 (Publication date: July 15, 1994) JP-A-6-103073 (Publication date: April 15, 1994) JP-A-6-168226 (publication date: June 14, 1994) Special Table 2000-511304 (Publication Date: August 29, 2000) JP-A-6-301546 (Publication date: October 28, 1994) “Communications of the ACM, 38: 49-57 (1995)” D. Heckerman, J. Breese, K. Rommelse, “Decision-theoretic Trouble shooting”

  Conventional diagnosis techniques include case-based reasoning, rule-based reasoning, and diagnostic methods using Bayesian networks. In addition, in order to make a diagnosis, input data is required, but this data is combined with what is automatically acquired by a sensor, etc., and what is acquired by the person inputting the observation results. It is common.

  And in the said nonpatent literature 1, the mechanism which diagnoses through effective action (observation and repair) in view of the cost of the observation and repair which a person performs based on a Bayesian network is proposed.

  The mechanism of Non-Patent Document 1 assumes a fixed general user. In other words, it is assumed that the user can always perform actions (observation and repair) proposed by the system. However, there are usually individual differences among users, and the presented actions cannot always be performed. Conversely, when presenting only actions that can be performed by any user, actions that can be performed only by a specific user will not be presented, no matter how effective.

  Further, the conventional diagnosis system is premised on that the user does not improve diagnosis knowledge and skills by using the system. That is, even if the user gains experience and can perform effective actions for diagnosis, the system does not present such actions.

  Patent Document 3 describes an optimum load leveling plan planning method and apparatus for determining work sharing on the basis of minimization based on required time and time unit of each work, ability of each time unit, and work allocation constraint conditions. Has been.

  However, Patent Document 3 does not consider individual differences in work ability when a person works. In addition, when a person works, a person's ability grows by gaining experience, but does not mention that the ability is observed and updated as appropriate. Further, when a person works, OJT (On the Job Training) for improving a person's ability may be performed, but the optimization is not aimed at improving the ability.

  Patent Document 4 describes an agent-based instruction system and method for providing individual education using an agent.

  However, Patent Document 4 is characterized in that the learning content provided from the teacher to the student is individually selected, and the agent cannot substitute that the student cannot. Moreover, the subject who performs the task is a student, and the task is not completed by the cooperative work of the system and the student.

  Patent Document 5 describes an artificial intelligence software shell for plant operation simulation in which observation data and a target value are input and plant control is performed based on a knowledge base.

  However, in Patent Document 5, input from a person is only a target value, and a person is not actively involved in task execution. Further, since only data that can be acquired from the machine is used as observation data, there is no room for human judgment to intervene in task execution, and the execution is limited to simple task execution.

  The present invention has been made in view of the above-mentioned problems, and its purpose is a factor capable of optimizing the work sharing between human and machine according to the ability of the user in the cooperative diagnosis of human and machine. An estimation device, a factor estimation program, a recording medium on which the factor estimation program is recorded, and a factor estimation method are realized.

  In order to solve the above problems, a factor estimation apparatus according to the present invention is a factor estimation apparatus that estimates a factor from a result generated in a system to be diagnosed, and each of a plurality of results that can occur in the system. Estimated knowledge record for associating one or more candidate factors and recording factor estimation knowledge information indicating the diagnosis path of factor estimation from each result to each factor corresponding to the result as knowledge of the network structure by conditional branching And an inference processing means for performing factor estimation processing for estimating a factor from a result based on the factor estimation knowledge information recorded in the estimation knowledge recording unit, and a belief that is an index indicating the certainty of inference leading to the factor In the process of performing the factor estimation process by the certainty factor calculation means for calculating the degree for each factor and the inference processing means, the factor estimation knowledge information In a process in which factor estimation processing is performed by the input control unit that acquires the feature amount corresponding to the included condition from the user and the inference processing unit, the feature amount corresponding to the condition included in the factor estimation knowledge information is Feature amount calculation means that is calculated based on inspection result data obtained from inspection related to the system, and user cost that is a cost required for the user to determine the feature amount corresponding to the condition and input it through the input control means A user cost recording unit that records the machine cost in association with the condition, and a machine cost record that records the machine cost, which is the cost required for the feature quantity calculation means to calculate the feature quantity corresponding to the condition, in association with the condition And a non-input condition that is a condition in which no feature quantity is acquired or calculated by the input control means and the feature quantity calculation means. Confidence prediction that calculates the difference between the certainty factor predicted value, which is the certainty factor in the state where the feature value is set as the target non-input condition, and the current certainty factor, and calculates the sum of the certainty factor predictive change amount The change amount calculation means and the user cost and the machine cost corresponding to the target non-input condition are acquired from the user cost recording unit and the machine cost recording unit, respectively, Dividing the certainty factor predicted change amount calculated by the certainty factor predicted change amount calculating unit, the input effect value calculating unit for calculating the input effect value of the target non-input condition, the certainty factor predictive change amount calculating unit, and the input Each non-input condition is set as a target non-input condition by the effect value calculation means, and an input effect value is obtained for each non-input condition. It is characterized by comprising condition selection means for determining a non-input condition.

  The factor estimation method according to the present invention is a factor estimation method in a factor estimation device that estimates a factor from a result generated in a diagnosis target system, and is generated in the system recorded in the estimated knowledge recording unit. Factor estimation for associating one or more candidate factors with each of a plurality of results to be obtained and indicating the diagnostic path of factor estimation from each result to each factor corresponding to the result as network structure knowledge by conditional branching In the process of performing factor estimation processing to estimate the factor from the result based on the knowledge information, a certainty factor calculating step for calculating the certainty factor for each factor, which is an index indicating the certainty of the reasoning leading to the factor, and the factor estimating knowledge information The feature quantity corresponding to the conditions included in the test is not acquired from the user, and the test obtained from the test related to the system From the uninput condition, which is a condition that is not calculated based on the result data, to the selected unentered condition of interest, the certainty factor predicted value that is the certainty factor in the state in which the feature amount is set and the current certainty factor Required for calculating a certainty factor predictive change amount, which is the sum of the factors, and determining and inputting a feature value corresponding to the target non-input condition. A user cost that is a cost is acquired from the user cost recording unit, a machine cost that is a cost required to calculate a feature amount corresponding to the target non-input condition is acquired from the machine cost recording unit, and the user cost and the The input effect that calculates the input effect value of the target uninput condition by dividing the certainty factor predicted change amount calculated in the certainty factor predictive change amount calculating step with the smaller machine cost In the calculation step, the certainty predictive change calculation step, and the input effect value calculation step, each non-input condition is set as the target non-input condition, and the input effect value is obtained for each non-input condition, and the input effect value is large. And a condition selection step for determining one or a plurality of non-input conditions.

  According to the above-described configuration or method, the factor estimating apparatus can calculate a certainty factor predicted value that is a certainty factor in a state in which a feature value is set in a target non-input condition, which is one of the conditions in which the feature value is not input, and the current time The user cost, which is the cost required for the user to determine and input the feature amount corresponding to the target non-input condition, and calculate the sum (confidence predictive change amount) for each factor, The input effect value of the target non-input condition calculated by dividing the feature amount corresponding to the target non-input condition by the smaller of the machine cost, which is the cost required to calculate based on the inspection result data, for each non-input condition One or a plurality of non-input conditions are determined from those having a large input effect value.

  Here, in addition to quantitative indicators based on numerical values such as temperature and area, the feature quantity includes “large”, “mid”, “small”, “Yes”, “No”, etc. It also includes qualitative indicators. In addition, the inspection result data is obtained from an inspection related to the system, for example, an inspection of the system or a processing object by the system. Further, the inspection result data may be stored in the recording device in advance, or may be acquired by inspection when it is necessary for the calculation.

  Therefore, the certainty factor when the feature value is obtained can specify the most efficient uninput condition with respect to the cost for obtaining the feature value. This makes it possible to acquire feature quantities in order from the non-input condition with high cost performance regardless of the diagnostic path for factor estimation. In addition, as the cost used in the calculation, the cost with which the user determines the feature value and the cost with which the machine determines the feature value is smaller. Thereby, whether the feature amount is acquired from the user or the machine can be performed simultaneously with the specification of the uninput condition for acquiring the feature amount.

  Therefore, in collaborative diagnosis between humans and machines, it is possible to optimize the division of work between humans and machines and the processing order of uninput conditions according to the user's ability (knowledge, skills, etc.), enabling efficient diagnosis Become.

  Furthermore, the factor estimation apparatus according to the present invention is adapted to indicate a degree to which the feature amount satisfies a corresponding condition based on the feature amount acquired by the input control unit or the feature amount calculated by the feature amount calculation unit. A degree-of-fit calculation means for calculating the degree of reliability, wherein the degree-of-confidence calculation means uses a value representative of a set of degrees of suitability for the conditions included in the diagnostic path as the degree of confidence.

  According to said structure, the representative value of the adaptability which shows the grade which satisfy | fills the condition which a feature-value respond | corresponds further is made into the reliability of each diagnostic path | pass (namely, factor). Therefore, the certainty factor can be easily calculated. Note that examples of the representative value of the fitness include a minimum value and an average value.

  Furthermore, in the factor estimation device according to the present invention, when the user cost is smaller than the machine cost, the condition selection unit causes the input control unit to present a question corresponding to a condition for acquiring a feature amount to the user. It is characterized by being.

  According to said structure, the feature-value which should be acquired can be further acquired from a user as an answer with respect to a question.

  Furthermore, the factor estimation device according to the present invention measures the time from when the input control means presents a question corresponding to the condition for acquiring the feature amount to the user until the user inputs the answer, User cost management means for determining a user cost according to the measured time and recording the determined user cost in association with the condition is provided.

  According to the above configuration, the user's ability can be measured by measuring how long it takes the user to input the feature amount. Based on this time, the user cost can be updated. As the user cost, the answer time [seconds] may be used as it is, but may be obtained by other conversion or the like.

  Therefore, it becomes possible to appropriately cope with the user's ability to improve by gaining experience.

  Furthermore, the factor estimation device according to the present invention is characterized in that the input effect value calculation means uses, as a user cost, a value obtained by subtracting a predetermined value from the user cost acquired from the user cost recording unit.

  According to the above configuration, since the user cost is set to be smaller than the actual value by a predetermined value, a question having a higher level than the actual level of the user is presented. For example, “(target level−current level) × coefficient for each input condition” can be used as the predetermined value.

  Therefore, the ability improvement can be promoted by making the user experience the work at the ability level higher than the actual level. Therefore, it is possible to realize OJT through diagnostic work.

  The factor estimation device according to the present invention is suitable for a process management device that estimates a factor from a result generated in a processing system that performs processing on an object.

  The factor estimation device may be realized by a computer. In this case, a factor estimation program for causing the factor estimation device to be realized by a computer by causing the computer to operate as each of the above-described means, and recording the same Computer-readable recording media are also within the scope of the present invention.

  As described above, the factor estimation apparatus according to the present invention associates one or more factor candidates with each of a plurality of results that may occur in the system system to be diagnosed, and responds to the result from each result. Based on the factor estimation knowledge information recorded in the estimation knowledge recording unit and the estimation knowledge recording unit for recording factor estimation knowledge information indicating the diagnosis path of factor estimation leading to each factor as knowledge of the network structure by conditional branching Inference processing means for performing factor estimation processing to estimate the factor from the result, confidence factor calculation means for calculating the certainty factor, which is an index indicating the certainty of the inference leading to the factor, and factor estimation by the inference processing means In the course of processing, the input control means for acquiring the feature amount corresponding to the condition included in the factor estimation knowledge information from the user, and the inference In a process in which factor estimation processing is performed by the processing means, a feature amount calculation unit that calculates a feature amount corresponding to a condition included in the factor estimation knowledge information based on inspection result data obtained from the inspection related to the system; A user cost recording unit that records a user cost, which is a cost required for a user to determine a feature value corresponding to a condition and input it via the input control unit, and the feature value calculation unit; The feature amount is acquired and calculated by the machine cost recording unit that records the machine cost, which is the cost required to calculate the feature amount corresponding to the condition, in association with the condition, and the input control unit and the feature amount calculation unit. The certainty factor predicted value that is the certainty factor in the state where the feature amount is set in the target non-input condition selected from the non-input condition that is not the current condition and the current A difference with reliability is calculated for each factor, and a certainty predictive change calculation means for calculating a predictive predictive change that is the sum thereof, and a user cost and a machine cost corresponding to the target non-input condition are set as the user cost. Obtained from the recording unit and the machine cost recording unit, respectively, and divided the certainty factor predicted change amount calculated by the certainty factor predictive change amount calculating unit with the smaller of the user cost and the machine cost, The input effect value calculation means for calculating the input effect value of the condition, the certainty factor predicted change amount calculation means, and the input effect value calculation means, with each non-input condition as a focused non-input condition It is a structure provided with the condition selection means which calculates | requires for every condition and determines one or several uninput conditions from the thing with a large input effect value.

  In addition, the factor estimation method according to the present invention associates one or more factor candidates with each of a plurality of results that can be generated in the diagnosis target system and recorded in the estimated knowledge recording unit. In the process of performing the factor estimation process for estimating the factor from the result based on the factor estimation knowledge information indicating the diagnosis path of the factor estimation leading to each factor corresponding to the result as the network structure knowledge by conditional branching, the factor is reached A certainty factor calculation step for calculating a certainty factor that is an index indicating the certainty of inference for each factor, and a feature amount corresponding to the condition included in the factor estimating knowledge information is not acquired from the user, and the above From the non-input condition that is not calculated based on the inspection result data obtained from the system-related inspection, to the selected non-input condition selected A certainty factor predictive change calculating step for calculating a certainty factor predictive change amount that is a sum of the difference between the certainty factor predicted value that is the certainty factor in the state where the feature amount is set and the current certainty factor, The user determines a feature amount corresponding to the target non-input condition, acquires a user cost that is a cost required for input from the user cost recording unit, and calculates a feature amount corresponding to the target non-input condition Is obtained from a machine cost recording unit, and the user cost and the smaller one of the machine costs are divided by the certainty predictive variation calculated in the certainty predictive variation calculating step. In each of the input effect value calculating step for calculating the input effect value of the target non-input condition, the certainty predictive change amount calculating step, and the input effect value calculating step, As a force condition each remarked non input conditions, prompted effect value for each non-input conditions, the method comprising the condition selection step of determining one or more non-input conditions from having a large input effect value.

  For this reason, the certainty factor when the feature amount is obtained can identify the most efficient uninput condition with respect to the cost for obtaining the feature amount. This makes it possible to acquire feature quantities in order from the non-input condition with high cost performance regardless of the diagnostic path for factor estimation. In addition, as the cost used in the calculation, the cost with which the user determines the feature value and the cost with which the machine determines the feature value is smaller. Thereby, whether the feature amount is acquired from the user or the machine can be performed simultaneously with the specification of the uninput condition for acquiring the feature amount.

  Therefore, in collaborative diagnosis between humans and machines, it is possible to optimize the division of work between humans and machines and the processing order of uninput conditions according to the user's ability (knowledge, skills, etc.), enabling efficient diagnosis The effect of becoming.

  An embodiment of the present invention is described below with reference to the drawings. In the present embodiment, a process management system applied to a production system having a printed circuit board production line will be described. However, the present invention is not limited to a printed circuit board production system. It can be applied to general management. In addition, the treatment process of the object is, for example, an industrial product production process, an industrial product, an agricultural product, or a raw material inspection process, a waste object (for example, factory waste, factory wastewater, waste gas, garbage, etc.) It means a processing process, an inspection process for waste objects, an inspection process for equipment, a recycling process, and the like.

(Production system configuration)
First, a printed circuit board production system (processing system) 1 to which the process management system according to the present embodiment is applied will be described with reference to FIG. The production line in the production system 1 includes each process (printing process, mounting process, reflow process, etc.) for manufacturing a printed circuit board. In the example shown in the figure, the production system 1 includes a printing device 11 that performs a solder printing process of pasting solder on a substrate, a mounting device 12 that performs a component mounting process of mounting electronic components on the substrate, and an electronic component on the substrate. A soldering device 13 for performing a reflow process for soldering and a process management device (factor estimating device) 10 for managing the production system 1. The printing device 11, the mounting device 12, and the soldering device 13 are arranged in this order from upstream to downstream in the product flow of the production system 1.

  Further, a printing inspection device 14 a is disposed in the vicinity of the printing device 11, a mounting inspection device 14 b is disposed in the vicinity of the mounting device 12, and a soldering inspection device 14 c is disposed in the vicinity of the soldering device 13. . The print inspection device 14 a is for inspecting the quality of the substrate processed by the printing device 11. The mounting inspection device 14b inspects the substrate processed by the mounting device 12. The soldering inspection device 14 c is for inspecting the substrate processed by the soldering device 13. Hereinafter, when it is not necessary to distinguish the print inspection device 14a, the mounting inspection device 14b, and the soldering inspection device 14c, they are simply referred to as the inspection device 14.

  In addition, the process management apparatus 10 performs overall management of the entire production system 1 and performs factor estimation processing and analysis processing described later. The process management apparatus 10 receives various information inputs and instruction inputs from a user as a production manager and performs various processes.

  The process management device 10, the printing device 11, the mounting device 12, the soldering device 13, the printing inspection device 14a, the mounting inspection device 14b, and the soldering inspection device 14c form a communication network by being connected to each other via a communication line. ing. Note that the communication network may have any form as long as the apparatuses can communicate with each other. For example, a form in which a LAN (Local Area Network) is formed is assumed.

  In addition to the process management device 10, a terminal device on which a user performs operation input is provided separately in a state connected to the communication network, and data input to the process management device 10 and various screen displays are performed by this terminal device. It is good.

  In the above example, the inspection device 14 is provided corresponding to each of the printing device 11, the mounting device 12, and the soldering device 13. However, in the production system 1, at least one inspection device 14 is provided. What is necessary is just to be provided. For example, if at least the soldering inspection device 14c is provided, it is possible to detect defects occurring in the final manufacturing result.

(Configuration of process control device)
Next, the configuration of the process management apparatus 10 will be described below with reference to FIG.

  The process management device (factor estimation device) 10 estimates a failure factor from a failure result generated in the diagnosis target system. As shown in the figure, the process management apparatus 10 includes a control unit 30, an inspection result input unit (inspection result input means) 40, an input unit 21, a display unit 22, an estimated knowledge recording unit 23, a question data recording unit 24, an inference. A process temporary storage unit 25, a process state database (inspection result recording unit) 26, a user profile recording unit (user cost recording unit) 27, and a machine cost recording unit 28 are provided.

  The input unit 21 receives an instruction input from the user, an information input, and the like, and includes, for example, a key input unit such as a keyboard and buttons, a pointing device such as a mouse, and the like. The display unit 22 displays various processing contents in the process management apparatus 10, and is configured by a display device such as a liquid crystal display device or a CRT (Cathode Ray Tube).

  The inspection result input unit 40 receives data related to the inspection result of the manufacturing process in the production system 1. The print result input unit 41, the mounting result input unit 42, the soldering result input unit 43, and the manufacturing apparatus history input unit 44. It has. The print result input unit 41 receives the inspection result from the print inspection apparatus 14a. The mounting result input unit 42 receives the inspection result from the mounting inspection device 14b. The soldering result input unit 43 receives an inspection result from the soldering inspection device 14c. The manufacturing apparatus history input unit 44 receives information regarding the manufacturing history from the printing apparatus 11, the mounting apparatus 12, and the soldering apparatus 13.

  The inspection result input unit 40 receives information on the inspection result from at least one of the printing device 11, the mounting device 12, the soldering device 13, the printing inspection device 14a, the mounting inspection device 14b, and the soldering inspection device 14c. It only has to be like this. For example, if only the inspection result data relating to the soldering result is received from the soldering inspection device 14c, the inspection result data relating to the defect occurring in the final manufacturing result can be acquired.

  The estimated knowledge recording unit 23 records factor estimated knowledge information. The factor estimation knowledge information is information for searching for a factor for each of a plurality of failure results, and is recorded as a causal network. That is, the estimated knowledge recording unit 23 associates one or more defect factor candidates with each of a plurality of defect results that may occur in the diagnosis target system, and each defect corresponding to the defect result. The factor estimation knowledge information indicating the diagnosis path of factor estimation leading to the factor as knowledge of the network structure by conditional branching is recorded. Details of the causal network will be described later.

  The question data recording unit 24 records the question information presented to the user in performing factor estimation as a question database. Each piece of question information included in this question database is linked to factor estimated knowledge information recorded in the estimated knowledge recording unit 23, and corresponds to a condition between nodes in the causal network. As will be described later, an answer to the question information is an input value of the input item, and is a feature amount set in association with a condition between nodes.

  The inference process temporary storage unit 25 stores inference process information obtained according to the progress of inference in the inference process performed in the inference unit 32.

  The process state database 26 is a database that records data (inspection result data) related to the inspection result of the manufacturing process in the production system 1 accepted by the inspection result input unit 40. That is, the process state database 26 is manufactured from the inspection result by the printing inspection device 14a, the inspection result by the mounting inspection device 14b, the inspection result by the soldering inspection device 14c, the printing device 11, the mounting device 12, and the soldering device 13. Information about the history is recorded.

  The user profile recording unit 27 determines a feature amount corresponding to the condition and associates a user cost, which is a cost required for inputting through the question input / output control unit 51, with the condition (ie, input item). Record. In the present embodiment, the user cost is a user ID (user identification information) for specifying a user and a user level indicating the level of the user's ability, as well as for each user in the form of a user profile (FIG. 11A). To be recorded. The user profile recording unit 27 is realized by a non-volatile recording medium, and may be, for example, a hard disk device or the like fixed to the process management device 10 like other recording units. An IC card or the like prepared for each may be used. When the user profile recording unit 27 is prepared for each user, the user ID can be omitted. As will be described later, when updating the user cost or the user level, a rewritable recording medium is used as the user profile recording unit 27.

  The machine cost recording unit 28 records a machine cost, which is a cost required for the feature amount calculation unit 35 to calculate a feature amount corresponding to the condition, in association with the condition.

  The estimated knowledge recording unit 23, the question data recording unit 24, the process state database 26, and the machine cost recording unit 28 are realized by a non-volatile recording medium such as a hard disk device. The inference process temporary storage unit 25 is realized by a work memory such as a RAM (Random Access Memory).

  The control unit 30 controls processing in the process management apparatus 10, and includes an input display control unit (input control means) 31, an inference unit 32, a knowledge conversion unit 33, a question generation unit 34, and a feature amount calculation unit (features). (Quantity calculation means) 35 is provided.

  The knowledge conversion unit 33 reads the factor estimation knowledge information recorded in the estimation knowledge recording unit 23 and generates a production rule as inference knowledge. The production rule, which will be described in detail later, is information in a data format suitable for causing a computer to perform inference processing for factor estimation.

  In response to a request from the inference unit 32, the question generation unit 34 reads out question information corresponding to each process of the inference process from the question data recording unit 24, and generates a question.

  In response to a request from the inference unit 32, the feature amount calculation unit 35 reads the inspection result data recorded in the process state database 26 and performs a statistical calculation or the like to calculate a required feature amount. That is, in the process in which factor inference processing is performed by the inference processing unit 61, the feature amount calculating unit 35 determines the feature amount corresponding to the condition included in the factor estimation knowledge information as the diagnosis target system or the processing product by the system. Calculation is based on inspection result data obtained by inspection. The inspection result data is acquired by the inspection result input unit 40 and stored in the process state database 26.

  The input display control unit 31 receives input information from the input unit 21 and performs display control on the display unit 22. The input display control unit 31 includes a question input / output control unit (input control unit) 51 and a factor output control unit 52. I have.

  The question input / output control unit 51 displays a question on the display unit 22 in response to an instruction from the inference unit 32, and performs a process of receiving an answer input to the question and transmitting it to the inference unit 32. That is, the question input / output control unit 51 acquires, from the user, the feature amount corresponding to the condition included in the factor estimation knowledge information in the process in which factor inference processing is performed by the inference processing unit 61.

  Here, in addition to quantitative indicators based on numerical values such as temperature and area, the feature quantity includes “large”, “mid”, “small”, “Yes”, “No”, etc. It also includes qualitative indicators. On the question screen, the feature quantity to be input may be displayed in the form of choices so that it can be selected from the screen. Note that the feature quantity of the input item is stored in the inference process temporary storage unit, but “Null” is stored when the feature quantity has not been entered. Thereby, it can be determined whether the feature amount of each input item has been input or not.

  The factor output control unit 52 displays on the display unit 22 candidate factors estimated in accordance with instructions from the inference unit 32 and information on each factor.

  The inference unit 32 performs an inference process for estimating a factor, and includes an inference processing unit (inference processing unit) 61, a suitability calculation unit (fitness calculation unit) 62, and a certainty factor calculation unit (confidence factor calculation unit) 63. , An input item selection unit (condition selection unit) 64, and a user profile management unit (user profile management unit, user cost management unit) 65.

  The inference processing unit 61 performs a process for comprehensively controlling the factor estimation process. Specifically, the inference processing unit 61 performs factor estimation processing for estimating a failure factor from a failure result based on factor estimation knowledge information recorded in the estimated knowledge recording unit 23.

  The suitability calculation unit 62 calculates a suitability for a connected node according to the feature amount when a feature amount for an input item associated with a specific node of the network is obtained. I do. In other words, the fitness level calculation unit 62 is based on the feature value acquired by the question input / output control unit 51 or the feature value calculated by the feature value calculation unit 35, and indicates the degree to which the feature value satisfies a corresponding condition. Calculate the degree.

  The certainty factor calculation unit 63 performs a process of calculating the certainty factor for each estimated factor based on the fitness obtained by the factor estimation process. That is, the certainty factor calculation unit 63 calculates a certainty factor, which is an index (score) indicating the probability (likelihood) of inference leading to a defect factor, for each defect factor (confidence factor calculation step). Note that the certainty factor calculation unit 63 may use a value representing a set of suitability levels for the conditions included in the diagnosis path as the certainty factor. In addition, the posterior probability, a score based on the posterior probability, or a score based on a distance such as the Euclidean distance may be used as the certainty factor.

  The input item selection unit 64 performs an input item selection process (condition selection step). Therefore, the input item selection unit 64 includes a certainty factor predicted change amount calculation unit (confidence factor predicted change amount calculation unit) 64a and an input effect value calculation unit (input effect value calculation unit) 64b.

  The certainty factor predictive change amount calculation unit 64a is characterized by the attention uninput condition selected from one of the non-input conditions that is a condition in which the feature amount is not acquired or calculated by the question input / output control unit 51 and the feature amount calculation unit 35. The difference between the certainty factor predicted value, which is the certainty factor in the state where the amount is set, and the current certainty factor are obtained for each failure factor, and the certainty factor predictive change amount that is the sum thereof is calculated (the certainty factor predictive change amount calculation step) ).

  The input effect value calculation unit 64b acquires the user cost and the machine cost corresponding to the target non-input condition from the user profile recording unit 27 and the machine cost recording unit 28, respectively, The certainty factor predicted change amount calculated by the certainty factor predictive change amount calculation unit 64a is divided to calculate the input effect value of the target non-input condition (input effect value calculating step).

  As described above, the input item selection unit 64 uses the certainty factor predicted change amount calculation unit 64a and the input effect value calculation unit 64b to determine each non-input condition as a focused non-input condition, and obtain an input effect value for each non-input condition. One or a plurality of non-input conditions are determined from those having a large input effect value. Then, when the user cost is smaller than the machine cost, the input item selection unit 64 causes the question input / output control unit 51 to present a question corresponding to the condition for acquiring the feature amount to the user. Further, when the machine cost is smaller than the user cost, the feature amount calculation unit 35 is caused to acquire necessary inspection result data from the process state database 26 and calculate the feature amount. When the user cost is equal to the machine cost, the feature amount may be obtained by any route, but may be obtained from the question input / output control unit 51 from the viewpoint of user education.

  The user profile management unit 65 performs user profile management processing. Specifically, the user profile management unit 65 reads and writes user cost data recorded in the user profile recording unit 27. In addition, the user profile management unit 65 includes a timer (not shown), and in the user cost update mode, which will be described later, the question input / output control unit 51 presents a question corresponding to a condition for acquiring the feature value to the user. The time from when the user inputs an answer is measured, the user cost is determined according to the measured time, and the determined user cost is recorded in association with the condition. Thereby, since the question according to the skill level of the user who inputs the feature amount can be presented, the work sharing between the person and the machine can be optimized for each user.

  Further, the input effect value calculation unit 64b uses a value obtained by subtracting a predetermined value from the user cost acquired from the user profile recording unit 27 as the user cost in the education mode described later. As this predetermined value (target reduction amount), for example, “(user level target−user level) × coefficient for each input item” can be used. Here, the user level is the current user level and can be acquired from the user profile. The user level target is a level that should be the target, and is higher than the current user level. The coefficient for each input item is a coefficient for converting a level difference (user level target-user level) common to each input item into a target reduction amount of the user cost of each input item. The input effect value calculation unit 64b is set in advance. As a result, the user cost is lower than the actual level, and a higher level question is presented than the actual level. The predetermined value may be determined according to a function or table that is set in advance for each input item and has a user level and a user level target as two variables.

  Details of the inference process, the fitness calculation process, the certainty calculation process, the input item selection process, and the user profile management process will be described later.

(Causal network)
Next, a causal network as factor estimation knowledge information recorded in the estimation knowledge recording unit 23 will be described. The causal network is information indicating an inference process from each failure result to a factor of the failure result (hereinafter referred to as a failure factor) as knowledge of the network structure. In the causal network, there are a plurality of nodes in the middle of a diagnostic path from a failure result to a failure factor. When a branch of a diagnostic path occurs at this node, a diagnostic path from a specific failure result to a plurality of failure factors is formed. The causal network may be information indicated as knowledge of a tree structure in which the parent node is at most one.

  Each node shows a specific phenomenon. When there is a high possibility that the cause of the phenomenon corresponding to a certain node is a phenomenon corresponding to a downstream node in the diagnostic path of the node, these two The degree of matching between nodes will be high.

  FIG. 3 schematically illustrates an example of a causal network. In the drawing, d1 indicates a failure result, c1 to c4 indicate failure factors, and s1 to s5 indicate phenomena corresponding to nodes. In the example shown in the figure, the failure factors c1 to c4 are assumed as candidates for the failure result of d1, and the diagnostic path corresponding to each failure factor is configured by a chain of nodes s1 to s5. . In the case of this causal network, five diagnostic paths are included as shown in FIG.

  Here, for example, when the knowledge structure is such that the failure factors c1 to c4 are directly connected to the failure result of d1, the conditions for determining the degree of fitness for each failure factor are extremely complicated. On the other hand, when the knowledge structure of the network structure as described above is used, the conditions for determining the conformity of the phenomenon causing the failure result, the phenomenon causing each phenomenon, and the failure factor causing each phenomenon are as follows. If you look at it, it will be relatively simple. Therefore, it is possible to perform reasoning inference processing by verifying relatively simple conditions, and even a user with a low level of skill can appropriately narrow down the factors.

(Generation of production rules)
Next, a process of generating a production rule from the causal network diagnosis path by the knowledge conversion unit 33 will be described. As described above, the production rule is information in a data format suitable for causing a computer to perform inference processing for factor estimation. The knowledge conversion unit 33 reads out information on the target diagnostic path from the causal network information recorded in the estimated knowledge recording unit 23, and generates a production rule corresponding to the diagnostic path.

  A production rule includes information on conditions between nodes during a specific diagnosis path from failure results to failure factors, and failure factor information that is true when all of these conditions are met. is there. Therefore, it is possible to determine whether or not the corresponding failure factor is true by making a simple determination of whether or not all the conditions between the nodes are satisfied, and the information is suitable for computer processing. Yes.

  FIG. 5A shows an example of a diagnostic path. In the example shown in the figure, the diagnosis path is a path from the failure result of d1 to the failure factor of c1 through the nodes as phenomena of s1 to s3 in this order. In addition, information regarding conditions regarding f1 to f4 is set as diagnosis knowledge corresponding to the conditions between the nodes. That is, in order to estimate that the defective result of d1 is derived from the phenomenon of s1, it is necessary to satisfy the condition of f1 = large. Similarly, the condition for estimating from s1 to s2 is f2 = Large, the condition for estimation from s2 to s3 is f3 = large, and the condition for estimation from s3 to c1 is f4 = large.

  Based on the diagnostic path information as described above, the knowledge conversion unit 33 generates a production rule as shown in FIG. The production rule is information that the corresponding failure factor is true when the condition obtained by combining the conditions between all the nodes included in the diagnosis path from the failure result to the failure factor with AND is satisfied. The production rule shown in FIG. 5B shows information that c1 is true when all the conditions of f1 = large, f2 = large, f3 = large, and f4 = large are satisfied. .

  In addition, when multiple conditions combined with "and" are set as conditions between nodes, the production rule includes the multiple conditions combined with "and" and diagnoses from failure results to failure factors. When the condition obtained by combining the conditions between all the nodes included in the path with “and” is satisfied, the information indicates that the corresponding failure factor is true.

  In the example shown in FIG. 6A, the conditions for the estimation from s1 to s2 are f2 = large and f5 = large. As shown in FIG. 6B, the production rule in this case is that c1 is true when all the conditions of f1 = large, f2 = large, f5 = large, f3 = large, and f4 = large are satisfied. It becomes information that it is.

  Further, when a plurality of conditions combined with or are set as conditions between nodes, production rules corresponding to the respective conditions combined with or are generated. In other words, each production rule is included in the diagnosis path from the failure result to the failure cause, and is combined with all the conditions between the nodes other than the nodes set with multiple conditions combined with or. The information is that the corresponding failure factor is true when the condition obtained by combining any one of the conditions with AND is satisfied. In addition, production rules are generated as many as the number of conditions connected by or.

  In the example shown in FIG. 7A, the condition for estimation from s1 to s2 is f2 = large or f5 = large. In this case, as shown in FIG. 7B, a production rule that c1 is true when all the conditions of f1 = large, f2 = large, f3 = large, and f4 = large are satisfied, Two of the production rules that c1 is true when all the conditions of f1 = large, f5 = large, f3 = large, and f4 = large are satisfied are generated.

  If a supplementary path exists, a production rule is generated by combining and combining the conditions included in the supplementary path with the production rule corresponding to the diagnostic path leading to the same failure factor. The node of the diagnostic path defines an intermediate phenomenon for reaching the cause from the defect. However, there is knowledge that describes phenomena directly related to factors such as overall trends, not intermediate phenomena. This is described as a supplementary path.

  In the example shown in FIG. 8A, there is a diagnostic path from the phenomenon of s4 to the failure factor of c1 as a supplementary path, and this condition is f5 = large. In this case, as shown in FIG. 8B, when all the conditions of f1 = large, f2 = large, f3 = large, f4 = large, and f5 = large are satisfied, c1 is true. A rule is generated.

(Fitness calculation process)
Next, the fitness calculation processing by the fitness calculator 62 will be described. The degree of conformity indicates a degree that satisfies the conditions set between the nodes in the production rule. This degree of conformity is represented by a number from 0 to 1, and the higher the number, the higher the condition is satisfied.

  Here, when the condition has only two choices of whether the condition is completely satisfied or not completely satisfied, the degree of conformity is two, 1 or 0. On the other hand, in the present embodiment, by using fuzzy theory, an answer to a condition where the fitness is a value larger than 0 and smaller than 1 is allowed. Thereby, even when the answer to the condition is ambiguous, an inference result reflecting the ambiguous state can be obtained. Therefore, it is not necessary to force an ambiguous answer to an extreme answer, and it is possible to prevent improper inference. Further, since the numerical value of the fitness level indicates the degree of certainty of inference to the next node, the numerical value of the fitness level can be used as a material for determining the suitability of the inference.

  Specifically, first, a membership function is set for each language value that is a candidate for an answer to the corresponding condition. The fitness for each language value is calculated by referring to the membership function according to the language value input as an answer. More specifically, first, the fitness for the language value input as an answer is 1.0. Then, the fitness for a language value other than the language value input as an answer is determined by taking the min between the membership function corresponding to the language value and the membership function corresponding to the input language value, and calculating the max value thereof. It shall be obtained by calculation. Here, when there is no intersection between the membership function corresponding to the language value input as an answer and the membership function corresponding to the input language value, the fitness is set to 0.0.

  Information about the language value that is a candidate for the answer to each condition and each membership function is recorded in the estimated knowledge recording unit 23. Then, the fitness calculator 62 calculates the fitness by reading out the information recorded in the estimated knowledge recording unit 23. In addition, the fitness level calculation unit 62 records the calculated fitness level information in the inference process temporary storage unit 25 together with information on conditions corresponding to the fitness level.

  Here, as an example, the condition “is is large?” Is set, and three types of responses “large”, “mid”, and “small” are assumed as answers to this. In the following description, the degree of matching calculation process will be described. FIG. 9A and FIG. 9B show an example of a membership function for three types of answers, “large”, “mid”, and “small”. In the example shown in FIG. 9A, if the answer is “large”, the fitness as large is 1.0, the fitness as mid is 0.5, and the fitness as small is 0. .0. In the example shown in FIG. 9B, if the answer is “large”, the fitness as large is 1.0, the fitness as mid is 0.7, and the fitness as small. Becomes 0.3.

(Confidence calculation process)
Next, the calculation process of the certainty factor by the certainty factor calculation part 63 is demonstrated. The certainty factor is a value provided corresponding to each failure factor, and is a value indicating a high possibility that the failure factor is a failure factor of a failure result as an estimation target. In other words, the certainty factor is an index indicating the certainty of inference leading to a failure factor as an inference result.

  The certainty factor is obtained as follows. First, a production rule corresponding to a diagnostic path from a failure result as an estimation target to a corresponding failure factor is extracted. Then, among the conditions included in the extracted production rule, the condition for which the fitness is calculated is extracted, and the fitness with the smallest value among these is set as the certainty. In addition, when there are multiple production rules corresponding to the diagnostic path from the failure result to be estimated to the corresponding failure factor, the certainty factor is calculated for each production rule, and among the calculated certainty factors The largest certainty factor is set as the certainty factor for the corresponding failure factor.

  Note that the certainty factor calculation unit 63 acquires the necessary goodness-of-fit by reading out the goodness-of-fit information recorded in the inference process temporary storage unit 25 and the condition information corresponding to the goodness-of-fit, Calculate the degree. The calculated certainty factor is recorded in the inference process temporary storage unit 25 together with information on the corresponding failure factor.

For example, in the case of the example shown in FIGS. 7A and 7B, the certainty factor for the failure factor of c1 is obtained as follows. First, the fitness for all conditions f1 to f5 is extracted. Here, these fitness levels are defined as g1 to g5. Then, the certainty factor for the failure factor of c1 is
certainty factor (c1) = max (min (g1, g2, g3, g4), min (g1, g3, g4, g5))
Sought by.

  In the above example, the degree of conformity that is the smallest value among the degrees of conformance of the conditions included in the extracted production rule is set as the certainty, but the present invention is not limited to this. Instead, it may be a value representative of a set of fitness included in the production rule, for example, an average value.

(Calculation process of certainty predictive change)
Next, the calculation process of the certainty degree predictive change amount by the certainty degree predictive change amount calculation unit 64a will be described. The certainty factor predictive change amount is a value provided corresponding to each input item (that is, a condition), and when the feature amount for the input item is obtained, the sum of the change amount of the certainty factor for each defect factor. Is a value indicating That is, the certainty prediction change amount is expressed by the following equation.
Confidence prediction change amount (s) = Σc | confidence prediction value (s) −current confidence |
Here, Σc represents the total sum of the failure factors. The certainty factor predicted value is a certainty factor in a state in which a feature amount is set in a focused non-input item selected from non-input items (focused non-input condition).

  Specifically, the certainty factor predicted change amount calculation unit 64a first reads the information of the current certainty factor for each failure factor recorded in the inference process temporary storage unit 25 to obtain the necessary certainty factor. get. FIG. 10 shows a specific example of the current certainty factor stored in the inference process temporary storage unit 25. As shown in the figure, the certainty factor is stored for each failure factor.

  Next, the certainty factor predictive change amount calculation unit 64a assumes that data has been obtained for the target uninput item, causes the goodness factor calculation unit 62 to calculate the goodness degree, and the certainty factor calculation unit 63 determines the certainty factor. Let the degree be calculated. This certainty factor is the above certainty factor predictive value. Then, the certainty factor predictive change amount calculation unit 64a calculates the certainty factor predictive change amount according to the above arithmetic expression. This is repeated for each uninput item. The calculated certainty factor predicted change amount is recorded in the inference process temporary storage unit 25 together with information on the corresponding input item (specifically, the item ID of the input item).

(Input effect value calculation processing)
Next, input effect value calculation processing by the input effect value calculation unit 64b will be described. The input effect value is a value provided corresponding to each input item (ie, condition), and is a value obtained by dividing the certainty prediction change amount by the cost required to obtain the feature amount corresponding to the target uninput item. It is. Then, the smaller of the user cost corresponding to the feature amount and the machine cost is used as the cost. That is, the input effect value is expressed by the following equation.
Input effect value (s) = confidence predictive change (s) / min (user cost (s), machine cost (s)).

  Specifically, the input effect value calculation unit 64 b first reads information on the certainty factor predictive change amount for each non-input item recorded in the inference process temporary storage unit 25. In addition, the input effect value calculation unit 64b determines the feature amount corresponding to the target non-input item by the user, reads out the user cost that is a cost required for the input from the user profile recording unit 27, and the target non-input item Is read from the machine cost recording unit 28. The machine cost is a cost required for the feature amount calculation unit 35 to calculate the feature amount corresponding to the inspection result data.

  Next, the input effect value calculation unit 64b compares the user cost and the machine cost, and calculates the input effect value by dividing the certainty factor predicted change amount by the smaller one. And the input effect value calculating part 64b calculates | requires an input effect value for every non-input item. The calculated input effect value is recorded in the inference process temporary storage unit 25 together with information on the corresponding input item (specifically, the item ID of the input item).

  Here, FIGS. 11A and 11B are tables showing examples of user profiles. As shown in the figure, the user cost is recorded in the user profile in association with the input item. 11A and 11B correspond to the diagnostic path shown in FIG.

  The user cost is, for example, the user's work time [seconds], more specifically, from the time when the question input / output control unit presents a question corresponding to the uninput item to the time when the user inputs an answer. Can be defined as time [seconds]. However, the present invention is not limited to this example.

  In this embodiment, it is assumed that the user cost decreases according to the user's ability. That is, the user cost for the expert is small, and the user cost for the beginner is large. The user level is assumed to increase according to the user's ability. That is, the user level of the expert is large and the user level of the beginner is small.

  FIG. 11B shows a state where the user cost has been lowered (s1, s3) from FIG. 11A and the user profile has been updated in the user cost update mode. In FIG. 11B, the user level is also updated.

  FIGS. 12A and 12B are tables showing examples of machine costs. As shown in the figure, the machine cost is recorded in association with the input item. FIGS. 12A and 12B correspond to the diagnostic path shown in FIG.

  The machine cost can be defined as, for example, the calculation time [seconds] of the feature amount calculation unit 35. However, the present invention is not limited to this example. For example, the cost for using the inspection apparatus may be reflected.

  FIG. 12A shows an example of a simple definition such that “0” [seconds] can be acquired by the calculation of the feature amount calculation unit 35 and “∞” [seconds] cannot be acquired. It is. In contrast, FIG. 12B is an example defined in detail.

In the educational mode, the input effect value is expressed by the following equation.
Input effect value (s) = confidence prediction change amount (s) / min (user cost (s) −target reduction amount (s), machine cost (s)).
Target reduction amount (s) = coefficient (s) × (user level target−user level).
Here, the coefficient (s) is a coefficient for converting a level difference (user level target-user level) common to each input item into a target reduction amount of the user cost of each input item. Specifically, the coefficient (s) is such that when the user's level is increased by the above level difference, a value corresponding to the time expected to reduce the response time of the input item to be obtained is obtained. Is set. The coefficient (s) is preset in the input effect value calculation unit 64b for each input item. Note that the target reduction amount (s) may be determined in accordance with a function or a table that is set in advance for each input item and has two variables for the user level and the user level target.

(Input item selection process)
Next, input item selection processing by the input item selection unit 64 will be described. Specifically, the input item selection unit 64 reads information on the input effect value for each non-input item recorded in the inference process temporary storage unit 25. Then, an uninput item having the maximum input effect value is selected. Note that one or a plurality of non-input items may be selected not only from the non-input item having the largest input effect value but also from those having a large input effect value.

  When the user cost corresponding to the selected non-input item is smaller than the machine cost, the input item selection unit 64 causes the question input / output control unit 51 to present a question corresponding to the non-input item to the user. At this time, in detail, the input item selection unit 64 sends to the inference processing unit 61 the input item ID of the selected uninput item and an instruction that the feature amount should be acquired from the question input / output control unit 51. Thereafter, the question is presented and the feature amount is acquired under the control of the inference processing unit 61.

  On the other hand, when the machine cost corresponding to the selected non-input item is smaller than the user cost, the input item selection unit 64 causes the feature amount calculation unit 35 to calculate the feature amount corresponding to the non-input item. At this time, in detail, the input item selection unit 64 sends to the inference processing unit 61 the input item ID of the selected uninput item and an instruction that the feature amount should be acquired from the feature amount calculation unit 35. Thereafter, the feature amount is acquired under the control of the inference processing unit 61.

  As described above, in the case of FIGS. 11A and 12A, when all the input items s1 to s5 are not input, the input effect value calculation unit 64b compares the user cost and the machine cost of each input item. The user items are used for the input items s1 to s3, and the machine cost is used for the input items s4 to s5. And when the input item selection part 64 acquires any feature-value of input items s1-s3, when acquiring any feature-value of input items s4-s5, it acquires from a user. Control is performed so that the feature amount calculation unit 35 calculates.

(Factor estimation process flow)
Next, the flow of factor estimation processing will be described with reference to the flowchart shown in FIG. When the factor estimating process is started, first, in step 1 (hereinafter referred to as S1), a question screen (FIGS. 17 and 18) is displayed on the display unit 22 by the input display control unit 31. In the question display area on the question screen, an area for inputting a defect result is provided. The failure result is input by the user to the area for inputting the failure result, and this input information is received by the question input / output control unit 51 (S2). Then, the failure result information is stored in the inference process temporary storage unit 25 by the inference processing unit 61.

  When the failure result is input, the inference processing unit 61 extracts the causal network information for the corresponding failure result from the estimated knowledge recording unit 23 (S3). At this time, the inference processing unit 61 acquires the causal network information converted into the production rules by the knowledge conversion unit 33. And the inference part 32 performs an inference process based on the extracted causal network (S4). After the inference process is completed, the factor output control unit 52 causes the display unit 22 to display a diagnosis result screen (FIGS. 17 and 18) (S5). Thereafter, the factor estimation process is terminated.

  Here, the inference process of the process management apparatus 10 has three modes (basic mode, user cost update mode, and education mode). Hereinafter, these three modes will be described in order.

(1) Basic Mode The flow of inference processing (S4) in the basic mode will be described with reference to the flowchart shown in FIG.

  When the inference process is started, first, the user profile management unit 65 reads a user profile from the user profile recording unit 27 (Sa1). Also, a list of input items is read from the estimated knowledge recording unit 23 (Sa2). Further, the data of the input item value (feature value) stored in the inference process temporary storage unit 25 is cleared (Sa3).

  Next, the certainty factor predicted change amount calculation unit 64a calculates the certainty factor predicted value of the non-input item, and calculates the certainty factor predicted change amount from the current certainty factor and the certainty factor predicted value of the uninput item (Sa4). ). Thereafter, the input effect value calculation unit 64b calculates the input effect value of each non-input item by dividing the certainty factor predicted change amount of the non-input item by the smaller user cost and machine cost (Sa5).

  Next, the input item selection unit 64 selects a non-input item having the maximum input effect value (Sa6). At this time, not only the largest uninput item but also one or a plurality of uninput items may be selected from those having a large input effect value.

  Then, based on the user cost and the machine cost of the selected non-input item, a feature amount acquisition destination is determined (Sa7). Specifically, when the machine cost is lower than the user cost, the machine is selected as a feature amount acquisition destination, and the process proceeds to Sa8. On the other hand, when the user cost is smaller than the machine cost, the user is selected as the feature amount acquisition destination, and the process proceeds to Sa10.

  When a machine is selected as a feature value acquisition destination in Sa7, the feature value calculation unit 35 acquires data necessary for calculating the feature value from the process state database 26 (Sa8), and calculates the feature value (Sa9). ).

  On the other hand, when the user is selected as the feature amount acquisition destination in Sa7, the question input / output control unit 51 reads a question sentence for acquiring the feature quantity from the question data recording unit 24 (Sa10), and the question sentence is read. It is presented to the user (Sa11), and a feature quantity is acquired as an answer to the question sentence (Sa12).

  Next, the inference processing unit 61 stores the calculation result of the feature amount calculation unit 35 or the user's answer result as the feature amount in the inference process temporary storage unit 25 in association with the input item (Sa13). Thereafter, the inference processing unit 61 performs inference (Sa14) and displays the inference result (Sa15). The processes of Sa4 to Sa15 are repeated, and when there are no uninput items, the inference process is terminated (Sa15).

  As described above, in the basic mode, based on the user cost and the machine cost for acquiring the feature value, the sharing of the person and the machine for obtaining the feature value and the processing order of the uninput items should be optimized. Can do. Therefore, it is possible to perform an efficient diagnosis according to the user's ability (knowledge, skill, etc.).

(2) User Cost Update Mode The flow of inference processing (S4) in the user cost update mode will be described with reference to the flowchart shown in FIG. The flow in the user cost update mode is different from the flow in the basic mode (FIG. 14) only in that Sb1 is added. Therefore, only the differences will be described.

  Sb1 is a step performed when a user is selected as a feature amount acquisition destination in Sa7. The user cost is updated from the answer result and answer time of the user, and is recorded in the user profile. This updated user cost is used in the next inference (dashed arrow in the figure).

  As a method for updating the user cost from the user's answer result and answer time, for example, the time (answer time) from when the question is presented until the user inputs the answer (answer time) is measured, and the answer time is determined from the value of the user cost. If it is small and the answer result is appropriate, the answer time is set as a new user cost. However, the present invention is not limited to this. For example, whether or not to update the user cost may be determined in consideration of the user's past answer history. Even when the user cost is updated, the answer time may not be adopted as it is, but may be selected from a preset value set in advance.

  As described above, in the user cost update mode, the user's ability is dynamically recognized and the user cost is updated by observing whether or not the user has been able to input the feature amount and how long it has been input. be able to. Therefore, it becomes possible to appropriately cope with the user's ability to improve by gaining experience.

(3) Education Mode The flow of inference processing (S4) in the education mode will be described with reference to the flowchart shown in FIG. The flow in the education mode is different from the flow in the user cost update mode (FIG. 15) only in that Sc1 is added and Sc2 is performed instead of Sa5. Therefore, only the differences will be described.

  Sc1 is a step performed at the beginning of the inference process, and reads a user level target. The user level target may be input by the user via the input unit 21 at the start of diagnosis, or may be read from the user profile recording unit 27 or the like.

  In Sc2, the input effect value is calculated using the cost obtained by reducing the user cost by the target reduction amount determined based on the user level and the user level target. Thereby, since the user cost is set smaller than the actual cost, a question at a higher level than the actual level is presented.

  As described above, in the education mode, not only the work efficiency but also the level of the user's ability (knowledge, skill, etc.) and the ability improvement effect by the work experience for the user's desired ability level are included. Each work assignment and work order can be optimized. Therefore, it is possible to realize OJT through diagnostic work.

(Display example of diagnosis result)
Next, display examples of diagnosis results will be described with reference to FIGS. 17 and 18. This diagnosis screen is provided with a question display area and a diagnosis result display area.

  In the question display area, a question presented to the user in the course of the factor estimation process and a history of the user's answer to the question are displayed. The question input / output control unit 51 displays the history by reading the history recorded in the inference process temporary storage unit 25.

  In the diagnosis result display area, a diagnosis path leading to the selected failure factor is displayed. This diagnostic path is extracted from the causal network by the inference processing unit 61, and the extracted diagnostic path is displayed in the diagnostic result display area by the factor output control unit 52.

  Also, on this diagnostic path, numerical information of the fitness level corresponding to the path between the nodes is displayed, and the corresponding path is displayed with a line thickness corresponding to the size of the fitness level. In addition, the display form of the fitness level is not limited to the above example, and only one of the fitness level information display and the path line thickness display may be displayed. In addition, the fitness level can be recognized by the user. Any display form may be used as long as it is in a form.

  In this way, by displaying the question answer history and diagnosis path information for a specific defect factor, it is possible to present the validity of the factor estimation result to the user.

  FIG. 17 shows an example of the diagnosis result for the failure factor c1 with respect to the failure result d1, and FIG. 18 shows an example of the diagnosis result for the failure factor “component contamination” with respect to the failure result of “bridge failure”. ing. As shown in these figures, the diagnosis result may be displayed even if the feature values of all the input items included in the diagnosis path are not obtained, and all the inputs included in the diagnosis path may be displayed. The diagnosis result may be displayed in a state where the feature amount of the item is obtained. Whether or not the feature amount of the input item is obtained depends on the input effect value of each input item.

(Specific examples of causal networks)
Next, a specific example of the causal network will be described with reference to FIG. In the example shown in the figure, a causal network for “bridge failure” as a failure result is shown. In this example, for “Bad defect”, “Mounting position deviation”, “Lead bending”, “Low paste flux activity”, “Part oxidation”, “Part dirt”, “Paste area” Eight defective factors are “large”, “paste misalignment”, and “heater temperature setting is high”. The diagnosis path from the failure result “bridge failure” to the above eight failure factors is set as knowledge of the network structure.

In the example shown in FIG. 19, the production rule obtained from the diagnostic path leading to the three failure factors of “mounting position misalignment”, “lead bend”, and “paste flux activity is low”
IF ((Contact between lead and land) = Yes & (Part displacement) = (Large)) then (Mounting displacement)
IF ((There is contact between the lead and land) = Yes & (Part displacement) = (Normal)) then (Lead bending)
IF ((There is a land without paste) = Yes & (There is non-wetting in the land) = Yes & (Paste flux activity) = (Low)) then (Paste flux activity is low)
It becomes.

In the example shown in FIG. 19, there are two diagnostic paths that lead to a failure factor of “the area of the paste is large”. Therefore, the following two production rules are obtained:
IF ((The heat dripping property is outside the specified value) = Yes & (Paste amount) = (Many)) then (Paste area is large)
IF ((There is a phenomenon that the paste wets to the shoulder of the lead) = Yes & (Reflow oven temperature) = (Normal)) then (Paste area is large)
It becomes.

In addition, in the example shown in FIG. 19, the diagnostic path leading to the failure factor “the heater temperature setting is high” includes the supplementary path.
IF ((There is a phenomenon in which the paste wets up to the shoulder of the lead) = Yes & (Reflow furnace temperature) = (High) & (Heater setting) = (High) & (A phenomenon in which the paste wets the entire board) Yes) then (the paste area is large)
It becomes.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims. That is, embodiments obtained by combining technical means appropriately changed within the scope of the claims are also included in the technical scope of the present invention.

  For example, in the above embodiment, a process management system for estimating a failure factor from a failure result in a production line has been described. However, the present invention is not limited to a “failure result” and a “failure factor”, and some “result” Therefore, it can be widely applied to diagnostic systems for estimating the “factor”. In addition, the present invention is not limited to a manufacturing process management system. For example, a failure diagnosis tool such as a computer system or a network system, software for supporting the handling of customer complaints, and a worker performing periodic inspections The present invention can be applied to various devices that estimate factors from the results, such as a system for assisting sales and a tool for supporting sales of customized products.

  The present invention includes means for holding inference conditions, means for generating information A necessary for inference under the above conditions, means for determining the collection destination and collection order of information A in consideration of efficiency, Means for obtaining information A from an external device (sensor, etc.), means for obtaining information A from a user, a capability (skill) level of the user, a means for holding whether information A can be collected, and a time required for collection; The inference system may include a means for holding a user's ability (skill) level, a means for executing an inference from the inference condition and information A, and a means for displaying an inference result.

  The present invention also includes means for holding inference conditions, means for generating information A necessary for inference under the above conditions, means for determining the collection destination and collection order of information A in consideration of efficiency, Means for obtaining information A from an external device (such as a sensor); means for obtaining information A from a user; means for holding whether information A is collected from a user; You may comprise as an inference system provided with the means to perform an inference from an inference condition and the information A, and the means to display an inference result.

  Further, the inference system may include a means for obtaining whether or not the information A is obtained from the user when the information A is obtained from the user, and a means for dynamically updating the ability level of the user. Good.

  Further, the inference system acquires means for holding a target for acquiring users, means for selecting one or more targets from among the targets for acquisition as targets B, In consideration of inference efficiency, the information A collection destination and the order for determining the collection order may be provided.

  Further, the inference system may present educational information for acquiring the target B.

  Further, the inference system may perform a production line diagnosis.

  Finally, each block included in the control unit 30 of the process management apparatus 10 may be configured by hardware logic, or may be realized by software using a CPU as follows.

  That is, the control unit 30 includes a CPU (central processing unit) that executes instructions of a control program that realizes each function, a ROM (read only memory) that stores the program, a RAM (random access memory) that expands the program, A storage device (recording medium) such as a memory for storing the program and various data is provided. An object of the present invention is to provide a recording medium on which a program code (execution format program, intermediate code program, source program) of a control program of the control unit 30 which is software that realizes the above-described functions is recorded in a computer-readable manner, This can also be achieved by supplying the control unit 30 and reading and executing the program code recorded on the recording medium by the computer (or CPU or MPU).

  Examples of the recording medium include a tape system such as a magnetic tape and a cassette tape, a magnetic disk such as a floppy (registered trademark) disk / hard disk, and an optical disk such as a CD-ROM / MO / MD / DVD / CD-R. Card system such as IC card, IC card (including memory card) / optical card, or semiconductor memory system such as mask ROM / EPROM / EEPROM / flash ROM.

  The control unit 30 may be configured to be connectable to a communication network, and the program code may be supplied via the communication network. The communication network is not particularly limited. For example, the Internet, intranet, extranet, LAN, ISDN, VAN, CATV communication network, virtual private network, telephone line network, mobile communication network, satellite communication. A net or the like is available. Also, the transmission medium constituting the communication network is not particularly limited. For example, even in the case of wired such as IEEE 1394, USB, power line carrier, cable TV line, telephone line, ADSL line, etc., infrared rays such as IrDA and remote control, Bluetooth ( (Registered trademark), 802.11 wireless, HDR, mobile phone network, satellite line, terrestrial digital network, and the like can also be used. The present invention can also be realized in the form of a computer data signal embedded in a carrier wave in which the program code is embodied by electronic transmission.

  The factor estimation apparatus according to the present invention is suitable for a process management system for estimating a defect factor in a production line, but is not limited to this, and supports a system for supporting customer complaint handling processing and a worker performing periodic inspection. The present invention can be applied to various devices that estimate factors such as systems and systems that support sales of customized products.

It is a block diagram which shows schematic structure of the process management apparatus which concerns on one Embodiment of this invention. It is a block diagram which shows schematic structure of the production system containing the said process management apparatus. It is a figure which shows an example of a causal network typically. It is a figure which shows the diagnostic path | pass obtained from the causal network shown in FIG. (A) is a figure which shows an example of a diagnostic path | pass, (b) is a figure which shows the production rule produced | generated by the diagnostic path | pass of (a). (A) is a figure which shows the other example of a diagnostic path | pass, (b) is a figure which shows the production rule produced | generated by the diagnostic path | pass of (a). (A) is a figure which shows the further another example of a diagnostic path | pass, (b) is a figure which shows the production rule produced | generated by the diagnostic path | pass of (a). (A) is a figure which shows the further another example of a diagnostic path | pass, (b) is a figure which shows the production rule produced | generated by the diagnostic path | pass of (a). (A) And (b) is a figure which shows an example of the membership function with respect to three types of answers of "large (large)", "normal (mid)", and "small (small)". It is a figure which shows the specific example of the present certainty degree memorize | stored in the inference process temporary storage part in the said process management apparatus in a table format. (A) and (b) are the figures which show the example of the user profile currently recorded on the user profile recording part in the said process management apparatus. (A) and (b) are the figures which show the example of the machine cost currently recorded on the machine cost recording part in the said process management apparatus. It is a flowchart which shows the flow of a factor estimation process. It is a flowchart which shows the flow of the inference process in basic mode. It is a flowchart which shows the flow of the inference process in user cost update mode. It is a flowchart which shows the flow of the inference process in education mode. It is a display screen figure which shows the example of the diagnostic screen which displays the diagnostic result with respect to a defect factor. It is a display screen figure which shows the specific example of the said diagnostic screen. It is a knowledge structure figure which shows the specific example of a causal network.

Explanation of symbols

10 Process management device (factor estimation device)
DESCRIPTION OF SYMBOLS 11 Printing apparatus 12 Mounting apparatus 13 Apparatus 14 Inspection apparatus 14a Printing inspection apparatus 14b Mounting inspection apparatus 14c Inspection apparatus 21 Input part 22 Display part 23 Presumed knowledge recording part 24 Question data recording part 25 Inference process temporary storage part 26 Process state database (inspection Results recording part)
27 User profile recording unit (user cost recording unit)
28 machine cost recording unit 30 control unit 31 input display control unit 32 inference unit 33 knowledge conversion unit 34 question generation unit 35 feature amount calculation unit (feature amount calculation means)
40 Inspection result input part (Inspection result input means)
41 Printing Result Input Unit 42 Mounting Result Input Unit 43 Soldering Result Input Unit 44 Manufacturing Equipment History Input Unit 51 Question Input / Output Control Unit (Input Control Unit)
52 factor output control unit 61 inference processing unit (inference processing means)
62 Goodness-of-fit calculation unit (goodness-of-fit calculation means)
63 certainty factor calculation unit (confidence factor calculation means)
64 Input item selection section (condition selection means)
64a Certainty factor predicted change amount calculation unit (confidence factor predicted change amount calculation means)
64b Input effect value calculation unit (input effect value calculation means)
65 User profile management unit (user cost management means)

Claims (8)

A factor estimation device for estimating a factor from a result generated in a diagnosis target system,
Each of a plurality of results that can occur in the system is associated with one or more candidate factors, and a diagnostic path for estimating a factor from each result to each factor corresponding to the result is represented by a network structure by conditional branching. An estimated knowledge recording unit that records factor estimated knowledge information indicated as knowledge;
Inference processing means for performing factor estimation processing for estimating a factor from a result based on the factor estimation knowledge information recorded in the estimated knowledge recording unit;
A certainty factor calculating means for calculating a certainty factor for each factor, which is an index indicating the certainty of inference leading to the factor,
In a process in which factor inference processing is performed by the inference processing means, input control means for acquiring a feature amount corresponding to a condition included in the factor inference knowledge information from a user;
In the process in which factor inference processing is performed by the inference processing means, feature amount calculation means for calculating a feature amount corresponding to the condition included in the factor estimation knowledge information based on inspection result data obtained from the inspection related to the system When,
A user cost recording unit that records a user cost, which is a cost required for a user to determine a feature amount corresponding to a condition, and input it via the input control unit;
A machine cost recording unit that records a machine cost, which is a cost required for the feature value calculation means to calculate a feature value corresponding to the condition, in association with the condition;
A certainty factor that is a certainty factor in a state in which a feature amount is set in a target non-input condition selected from one of the non-input conditions that is a condition in which the feature amount is not acquired or calculated by the input control unit and the feature amount calculation unit A difference between the predicted value and the current certainty factor is obtained for each factor, and a certainty factor predictive change amount calculating means for calculating a certainty factor predictive change amount that is a sum thereof,
User cost and machine cost corresponding to the target non-input condition are acquired from the user cost recording unit and the machine cost recording unit, respectively, and the certainty factor predictive change calculation is performed with the smaller of the user cost and the machine cost. Input effect value calculation means for calculating the input effect value of the target non-input condition by dividing the certainty factor predicted change amount calculated by the means;
By using the certainty factor predicted change amount calculating means and the input effect value calculating means, each non-input condition is set as a target non-input condition, and an input effect value is obtained for each non-input condition. A factor estimation device comprising: condition selection means for determining a non-input condition. Based on the feature quantity acquired by the input control means or the feature quantity calculated by the feature quantity calculation means, comprising: a fitness calculation means for calculating a fitness indicating the degree to which the feature quantity satisfies a corresponding condition;
The factor estimation apparatus according to claim 1, wherein the certainty factor calculation means uses a value representing a set of suitability levels for the conditions included in the diagnosis path as a certainty factor.   2. The condition selection unit, when the user cost is smaller than the machine cost, causes the input control unit to present a question corresponding to a condition for obtaining a feature amount to the user. Or the factor estimation apparatus according to 2;   The time from the time when the input control means presents a question corresponding to the condition for acquiring the feature amount to the user until the time when the user inputs the answer is measured, and the user cost is determined according to the measured time. 4. The factor estimating apparatus according to claim 3, further comprising user cost management means for recording the determined user cost in association with the condition.   The factor estimation according to any one of claims 1 to 4, wherein the input effect value calculation means uses a value obtained by subtracting a predetermined value from the user cost acquired from the user cost recording unit as the user cost. apparatus.   A factor estimation program for realizing the factor estimation device according to any one of claims 1 to 5 by a computer and causing the computer to function as each of the above means.   A computer-readable recording medium in which the factor estimation program according to claim 6 is recorded. A factor estimation method in a factor estimation device that estimates factors from a result generated in a diagnosis target system,
One or more candidate factors are associated with each of a plurality of results that can be generated in the system recorded in the estimated knowledge recording unit, and factor estimation from each result to each factor corresponding to the result is performed. In the process of performing the factor estimation process that estimates the factor from the result based on the factor estimation knowledge information indicating the diagnostic path as knowledge of the network structure by conditional branching,
A certainty factor calculating step for calculating a certainty factor for each factor, which is an index indicating the certainty of inference leading to the factor,
The feature quantity corresponding to the condition included in the factor estimation knowledge information has not been acquired from the user and is not input based on the condition that is not calculated based on the inspection result data obtained from the inspection related to the system The difference between the certainty factor predicted value that is the certainty factor in the state in which the feature amount is set in the selected uninput condition from the condition is obtained for each factor, and the certainty factor prediction that is the sum thereof A certainty factor predictive change amount calculating step for calculating a change amount;
The user decides the feature quantity corresponding to the focused non-input condition and acquires the user cost, which is the cost required for input, from the user cost recording unit, and calculates the feature quantity corresponding to the focused non-input condition Is obtained from the machine cost recording unit, and the user cost and the smaller one of the machine costs are divided by the certainty predictive change calculated in the certainty predictive change calculating step, An input effect value calculation step for calculating an input effect value of the target non-input condition;
In the certainty predictive change calculation step and the input effect value calculation step, each non-input condition is set as a target non-input condition, an input effect value is obtained for each non-input condition, And a condition selection step for determining a plurality of uninput conditions.

Images