Job Risk Assessment System, Model Generation Device, Job Risk Assessment Method, Job Risk Assessment Program Product

Abstract

The present invention provides an operation risk assessment system, a model generation device, an operation risk assessment method, and an operation risk assessment program. The operation risk assessment system includes: a model storage unit that stores an agent model, the agent model takes a first feature quantity of an operation as an input and outputs a second feature quantity of the operation calculated through a predetermined simulation of a process, the agent model is a learned model of the relationship between the input and the output, and its agent has a predetermined simulation with uncertainty, where uncertainty means that for each input, the output corresponding to the same value of the input is different; a prediction unit that repeatedly executes a process of taking the first feature quantity as an input and obtaining the second feature quantity as an output from the agent model for a predetermined number of trials for the same value of the first feature quantity, thereby predicting a plurality of second feature quantities with uncertainty for the same value of the first feature quantity in the process; and a risk assessment unit that performs a risk assessment of the operation in the process based on the plurality of second feature quantities with uncertainty predicted by the prediction unit.

Description

Technical Field

[0001] The present invention relates to an operation risk assessment system, a model generating device, an operation risk assessment method, and an operation risk assessment program product. Background Art

[0002] There is a system that evaluates a work plan consisting of multiple processes based on future forecasts. For example, there is a simulation system that uses current performance information and future forecast information to predict the future trends of a project consisting of multiple processes, performs a risk assessment of the project based on the forecast results, and presents the evaluation results to the user (see Patent Document 1).

[0003] Furthermore, in recent years, supply chains have been constructed across diverse systems and multiple organizations. In such situations, if productivity in any process decreases and delays occur due to issues such as manual labor, system incompatibility, or equipment failure, the overall work plan needs to be revised.

[0004] Furthermore, there is a monitoring system that simulates future trends that will be affected by a revision of a work plan and presents the simulation results to a user (see Non-Patent Document 1).

[0005] In these prior arts, future trends are simulated assuming that each process progresses at a standard pace.

[0006] Here, there is uncertainty: delays in individual processes due to various issues can affect subsequent processes, potentially delaying the entire work plan. However, the aforementioned prior art fails to account for this uncertainty, making it difficult to quickly simulate and assess the risk of work plans that complicate the model due to fluctuations in uncertainty within a realistic computational timeframe. Furthermore, the consideration of fluctuations in uncertainty makes it difficult to present the risk assessment results of work plans, derived from simulations involving complex models, to the user in an intuitive and understandable manner.

[0007] Patent Literature

[0008] Patent Document 1: Japanese Patent Application Laid-Open No. 2004-192109

[0009] Non-patent literature

[0010] Non-patent document 1: "実典から学ぶ|経嶶ダッシュボード成の记に気さえるべきポイント" , [online], Fanruan Technology Co., Ltd., [Retrieved on March 10, 2021], Internet<https: / / www.finereport.com / jp / analysis / northstarmetrics / > Summary of the Invention

[0011] The present invention has been made in view of the above circumstances, and an object thereof is to quickly perform a simulation and a risk assessment that take into account the uncertainty caused by problems in the risk assessment of an operation composed of processes.

[0012] In order to solve the above problems, in one aspect of the present invention, an operation risk assessment system performs a risk assessment of an operation composed of processes, and has: a model storage unit that stores an agent model that takes a first feature amount of the operation as an input and outputs a second feature amount of the operation calculated by a predetermined simulation of the process, and the agent model is a learned model of the relationship between the input and the output, and it substitutes the predetermined simulation with uncertainty, where the uncertainty means that for each input, the output corresponding to the same value of the input is different; a prediction unit that predicts a plurality of the second feature amounts having uncertainty for the same value of the first feature amount in the process by performing a process of taking the first feature amount as an input and obtaining the second feature amount as an output from the agent model by repeating a predetermined number of trials for the same value of the first feature amount; and a risk assessment unit that performs a risk assessment of the operation in the process based on the plurality of the second feature amounts having uncertainty predicted by the prediction unit.

[0013] According to the present invention, for example, in the risk assessment of an operation composed of processes, it is possible to quickly perform a simulation and a risk assessment that take into account the uncertainty caused by problems. Brief Description of the Drawings

[0014] Figure 1 An example of an operation performed through a plurality of processes having a hierarchical structure is shown.

[0015] Figure 2 An example of the progress of an operation having uncertainty is schematically shown.

[0016] Figure 3 An example of a prediction simulation of an operation schedule performed using two methods in the embodiment is schematically shown.

[0017] Figure 4 Examples of problem events are shown.

[0018] Figure 5 Examples of the input and output of the agent simulation of each process are shown.

[0019] Figure 6 The structure of the entire system of the embodiment is shown.

[0020] Figure 7It is a flowchart showing an example of the surrogate model generation process in the prior stage.

[0021] Figure 8 It is a flowchart showing an example of the risk analysis process in the application stage.

[0022] Figure 9 An example of the terminal display of the dashboard showing the risk analysis results.

[0023] Figure 10 An example of the terminal display of the dashboard showing the risk analysis results.

[0024] Figure 11 An example of the terminal display of the dashboard showing the risk analysis results.

[0025] Figure 12 It is a hardware diagram showing an example of the structure of a computer. Detailed implementation mode

[0026] Hereinafter, the embodiments of the present invention will be described using the text or drawings. However, the specific items of the structure, processing, data, and the number of each element shown in the present invention are not limited to the embodiments given here, and can be appropriately combined and improved within the scope of not changing the gist. In addition, elements not directly related to this embodiment are omitted from the illustration.

[0027] In the following description, symbols with subscripts are used to distinguish the same or similar constituent elements, and the main body of the symbol without the subscript is used as the general name of the same or similar constituent elements.

[0028] (Multiple processes with a hierarchical structure)

[0029] Figure 1 An example of an operation (for example, an operation in warehouse business, an operation in product manufacturing business, etc.) performed through multiple processes with a hierarchical structure is shown. As Figure 1 shown, each operation is performed in the order of SCM (Supply Chain Management) processes S(1), S(2), S(3), S(4). For example, in the SCM process S(2), each operation is performed in the order of the lower-level processes P(1), P(2), P(3), P(4), P(5). And, for example, in the process P(3), each operation is performed in the order of the lower-level operation processes M(1), M(2), M(3). The output of each process is output to the subsequent process. In Figure 1 , the lower-level processes are shown for the SCM process S(2) and the process P(3) respectively, but the same applies to other SCM processes and processes. Hereinafter, the process P(n) (n = 1, 2, 3, 4, 5) will be taken as an example for explanation.

[0030] (Progress of tasks with uncertainties)

[0031] Figure 2 An example of the progress of tasks with uncertainties is schematically shown. A task consisting of one or more processes is carried out according to an optimized task schedule. The task schedule includes: resource allocation of task equipment and task personnel generated based on a task plan including predicted values, and the task start times of each task equipment and task personnel, etc.

[0032] However, in practice, problems sometimes occur in each process, resulting in deviations from multiple patterns of the task schedule (advance and delay of the actual task progress, Figure 2 the dotted-line enclosed part of the schedule deviation). These deviations of multiple patterns become the uncertainties (indeterminacies) of the task progress (task time required, task end time, etc.). Among the deviations from the task schedule, the schedule delay in which the task time becomes longer than the task schedule becomes a problem.

[0033] In general process management, a margin buffer is set at each process stage, and scheduling management is performed to absorb errors within a determined range within the buffer. Sometimes, only when the plan deviation caused by the delay exceeds the buffer range, it is judged as a scheduling delay, and the task schedule is finely optimized by dynamically managing the margin buffer.

[0034] Here, generally, the schedule delay in each process is related to the previous and subsequent processes. That is, the schedule delay generated in a certain process sometimes affects the subsequent process, causing a schedule delay in the subsequent process. Moreover, not only the immediately following process, but sometimes it also affects more subsequent processes, causing a chain of schedule delays. Due to such a chain risk, it is necessary to re-evaluate the task schedule itself.

[0035] In order to avoid this chain risk in advance, it is desirable to perform agent simulation considering the occurrence of various problems that may be the main causes of the schedule delay in each process of the task schedule, and perform future prediction including the task schedule deviation. However, when considering the uncertainty of which problem among multiple problems will occur and making various changes to perform future prediction of the task schedule through agent simulation, the amount of calculation becomes huge, so there is a problem of insufficient computing speed of the computer and unable to perform such calculations within a practical time.

[0036] (Simulations using two methods)

[0037] Therefore, in the present embodiment, to solve the above problem of calculation time, two simulation methods, namely, the agent simulation of the upper layer and the agent simulation of the lower layer in the hierarchical simulation structure, are used. Figure 3Schematically shows an example of a predictive simulation of an operation schedule performed using two methods in an embodiment. In this embodiment, processes P(1) to P(5) are upper-level processes, and operation processes M(1) to M(3) are lower-level processes.

[0038] Agent simulation can represent action logic and internal states in a form understandable by humans, and can simulate various internal states of each process P(n) and their transitions due to the actions and interactions of agents under predetermined constraints.

[0039] Specifically, in agent simulation, the operation is decomposed into elements, and the required time for the elements is allocated along the time axis, thereby calculating the operation time. In addition, sometimes simulation calculations of operation accuracy and failure rate are performed considering randomness based on the relationship of the operations.

[0040] In this way, in agent simulation, some of the various phenomena or problems that occur during operation can be reproduced. For example, the waiting queue generated due to the overlap of the operation times of multiple agents, the conditions of the operation target object calculated physically, the required accuracy, and the operation difficulty are used to calculate the operation failure rate, etc., and the difference in the required time generated as a result is calculated. However, in agent simulation, the more diverse the problems that occur during operation, the more difficult it is to complete the prediction within a practical calculation time.

[0041] On the other hand, agent simulation uses a surrogate model of each process P(n) to proxy the calculation of agent simulation, and can simulate the same calculation results at high speed, where the surrogate model learns the output of agent simulation for the input.

[0042] When the agent model includes probabilistic indefinite elements (such as random variables) in the actions or interactions of the agents, it probabilistically acts on the same input and returns outputs of multiple patterns. For example, when the task input to process P(n) has specific characteristic quantities (such as the operation start time, product accuracy, etc.), the calculation of using the same task as the input of the agent model is repeated. Then, as the processing results of the same task, characteristic quantities of multiple patterns of the number of repetitions are output (operation end time, operation result accuracy, etc.). When the characteristic quantities of these multiple patterns become the input of the next process P(n + 1), the fluctuations of the characteristic quantities (operation end time, operation result accuracy, etc.) are maintained.

[0043] For example, as Figure 3As shown, consider the case where the proxy model for process P(3) is fed with input 1: assumptions, input 2: work start time (a time series of basic values), and one of the multiple patterns of work start time generated from the error model. In this case, the proxy model outputs output 1: internal state transitions and multiple (K(n+1) types) of output 2: work end time (a time series) based on the inputs. By repeatedly performing the processing from input to output on K(n) types of inputs, the proxy model can obtain multiple patterns of outputs corresponding to the multiple patterns of inputs.

[0044] Fluctuations in characteristic quantities (such as work completion time and work result accuracy) that serve as inputs and outputs for each process can be stored as a list of cases consisting of multiple pattern values or a probability distribution model. In this embodiment, the following set is used to represent fluctuations in characteristic quantities (such as work completion time and work result accuracy) for a single input. This set is a collection of multiple pattern results obtained through computational processing using a proxy model that includes probabilistic action elements.

[0045] For example, methods for implementing learned models containing such probabilistic action elements using Bayesian neural networks are known. In conventional neural networks, as a result of machine learning, the same output is always returned for the same input. In Bayesian neural networks, the coupling coefficient is represented as a distribution rather than a numerical value, returning different output results for the same input. This characteristic makes it possible to reproduce behavior (probability distribution) that causes the results to fluctuate probabilistically for the same input.

[0046] However, in this embodiment, as long as the method can simulate the various internal states and transitions of each process P(n) in the same way as agent simulation, it is not limited to agent simulation. In addition, as long as the learning method and the learned model can return multiple outputs with behaviors based on probabilistic distributions for the same input, it is not limited to Bayesian neural networks.

[0047] (Example of a problem event)

[0048] Figure 4 This represents an example of a problem event. In this embodiment, a typical "problem event" pattern is specifically predefined. A "problem event" refers to an internal state corresponding to the following state variable, which is pre-specified as a risk of causing problems such as decreased productivity during the execution of process P(n). When learning the agent model, the state of the state variable for each process that corresponds to the problem event in the agent simulation under predetermined assumptions is flagged and learned.

[0049] Specifically, information is maintained for grouping characteristic states in the behavior of each process P(n) into "problem events" for management. One or more problem events Qn,j (j = 1, 2, ……) are maintained for each process P(n). n is an index for identifying the process, and j is an index for identifying problem events within the same process P(n).

[0050] As Figure 4 shown, in the problem event table 17T, as the definition of a problem event, the process name 171, problem event name 172, nickname 173, case retrieval link 174, and pointer 175 to the discovery measurement function are stored in correspondence. The nickname 173 represents information about the problem event in a form understandable by humans. The case retrieval link 174 stores the retrieval link to the specific case of the problem event in the agent simulation. The pointer 175 to the discovery measurement function stores the pointer to the process function used to measure the discovery of the problem event in the agent simulation. The information form of the problem event is not limited to the tabular form.

[0051] Hereinafter, an example of a problem event in the operation management of a factory or warehouse is listed. Among the problem events, there are cases that occur due to input reasons such as input quantity, and cases that occur due to completely accidental probabilities.

[0052] · Reduction in productivity due to conveyor congestion

[0053] · Generation of temporary avoidance operations due to the overflow of the operation buffer set to absorb operation schedule delays

[0054] · Delay by more than a predetermined time (e.g., 15 minutes) from the scheduled time

[0055] · The required time until the final shipping time is less than the predetermined time (e.g., 30 minutes)

[0056] · Delay in operation productivity due to temporary resource shortages

[0057] · Return to the previous process for re-operation due to object damage

[0058] · Partial equipment local stop caused by equipment wear

[0059] (Input and output of the agent simulation for each process)

[0060] Figure 5 Examples of the input and output of the agent simulation for each process are shown. As Figure 5As shown, the information of the tasks input to process P(n) (n = 1 to 5) is held as time-series data (time migration model) An,k (k = 1, 2, ……). Each process P(n) has variables of internal states (s1, s2, s3, ……), and the productivity changes according to these variables. The task Ta1 input to the surrogate model for predicting the operation result of process P(1) is single, but the task Tan input to process P(n) (n = 2 to 5) is a set of K(n) types (K(n) > 1) of patterns. This is because state fluctuations occur due to the uncertainty during the processes before process P(n) (n = 2 to 5).

[0061] As candidates for the tasks input to each surrogate model for predicting the operation results of processes P(n) (n = 2 to 5) respectively, there are K(n) types of tasks Tan. Each task is a task representing a series of batch operations, and is time-series data reflecting the start time and state of each operation as time passes. And one task input to process P(n) is randomly sampled from the tasks Tan. By performing the surrogate simulation (neural network processing of the surrogate model) of process P(n) multiple times for one sampled data, different behavior cases can be obtained little by little.

[0062] In this way, as the output of process P(n) (n = 1 to 5), data of K(n + 1) types are output by adding the fluctuations within process P(n) to the data sampled from the input task Tan. Generally, K(n + 1) ≥ K(n). The obtained output data is used as the input data for the next process P(n + 1).

[0063] In addition, as a result of the surrogate simulation of process P(n), the migration of the internal state of process P(n) is also output. It is implied that in terms of performing the processing of process P(n), problems such as deterioration of productivity may occur due to changes in the state variables representing the internal state. The internal state represented by the state variables specified in advance in such an internal state is the above-mentioned "problem event".

[0064] (Structure of the entire system S)

[0065] Figure 6 Shows the structure of the entire system S of the embodiment. In this embodiment, as the entire system S, an example of a system that applies a system for performing risk analysis of the operation schedule in the operation environment E and visualizing the risk analysis results is shown, with the system for controlling the operation environment E such as a factory or a warehouse as the object.

[0066] The working environment E includes a working area, working equipment, and working personnel for performing operations on objects in each process. Working equipment and working personnel are allocated to each process. In the working environment E, according to the execution order of the operations of processes P(1) to P(5)( Figure 1 ), the working equipment (process P(1)) 40-1, the working equipment (process P(2)) 40-2, the working equipment (process P(3)) 40-3, the working equipment (process P(4)) 40-4, and the working equipment (process P(5)) 40-5 are connected in series via conveyors 50 (50-1, 50-2, 50-3, and 50-4).

[0067] The working equipment 40 represents working equipment including the personnel performing the operations of each process P(n) (n = 1 to 5). In Figure 6 , it is shown that the operation of process P(2) is performed by any one of the multiple working equipment (process P(2)) 40-2 connected in parallel. The same applies to process P(4) and the working equipment (process P(4)) 40-4.

[0068] The entire system S is configured to include a control system 1, a planning system 2, a control log storage unit 3, a risk assessment system 10, a simulation log storage unit 15, an agent model storage unit 16, a problem event storage unit 17, and a terminal 18. The control system 1, the planning system 2, the control log storage unit 3, and the risk assessment system 10 are connected in a communicable manner via a network N.

[0069] The control log storage unit 3, the simulation log storage unit 15, the agent model storage unit 16, and the problem event storage unit 17 are storage areas such as databases. The control log storage unit 3 holds the execution logs of the processes and controls executed by the control system 1 and the planning system 2. The simulation log storage unit 15 stores the execution results of the simulations performed by the agent simulation execution unit 11 and the prediction unit 13 for statistical analysis such as risk assessment. The agent model storage unit 16 stores the agent model 16M. The problem event storage unit 17 stores the problem event table 17T( Figure 4 ) and various associated information.

[0070] The terminal 18 is a computer for managers such as a tablet terminal having a touch panel and a display, which is connected to the risk assessment system 10 via a wireless communication line or a wired communication line.

[0071] The control system 1 and the planning system 2 constitute a job scheduling instruction system such as MES (Manufacturing Execution System) or WCS (Warehouse Control System). The control system 1 outputs job instructions to each job device 40 in real time according to the job schedule calculated by the planning system 2 to perform control. The planning system 2 calculates a job schedule indicating the optimal steps for performing jobs in processes P(1) to P(5).

[0072] The risk assessment system 10 simulates the jobs executed by the control system 1 according to the job schedule and evaluates the risks of the jobs. The risk assessment system 10 includes an agent simulation execution unit 11, an agent model generation unit 12, a prediction unit 13, and a risk assessment unit 14. The risk assessment system 10 is connected to a console (not shown) that receives the operations of the manager and outputs the status and results of the processing.

[0073] The agent simulation execution unit 11 executes a simulation of the transition of the internal state of each process presented by the behavior and interaction of agents in each process under predetermined assumption conditions.

[0074] The agent model generation unit 12 generates an agent model 16M used by the prediction unit 13 in a prior stage so as to be able to imitate the behavior of the agent simulation execution unit 11 at high speed. An agent model 16M is generated for each process.

[0075] The agent model generation unit 12 stores the action results obtained by randomly assigning various data to the agent simulation execution unit 11 in the simulation log storage unit 15. As the data of the action results, there are parameters of equipment operation conditions, the required time for each task (delay caused by the job), the execution accuracy of each task, etc.

[0076] The agent model generation unit 12 learns the agent model 16M including the action results and the order of the action results in order to learn the internal state transition model. In addition, the internal state equivalent to the "problem event" is also learned together, and the agent model 16M can output a judgment result of the occurrence of the problem event.

[0077] Regarding the generation of the agent model 16M, learning takes a long time, so the generation of the agent model is executed in a prior stage before the actual time operation of the entire system S (using the system as a CPS (Cyber-Physical System)). The prediction unit 13 uses the agent model 16M to execute the agent calculation simulation of the agent simulation execution unit 11 during actual time operation.

[0078] The risk assessment department 14 conducts a risk assessment of the evaluation object operation based on the processing results of the prediction department 13 and sends the evaluation results to the terminal 18.

[0079] (Agent model generation process)

[0080] Figure 7 It is a flowchart showing an example of the agent model generation process in the prior stage. The agent model generation process is executed by the agent model generation department 12 that has received the manager's instructions.

[0081] First, in step S11, the agent model generation department 12 randomly assigns data to the agent simulation execution department 11 to execute the agent simulation. Next, in step S12, the agent model generation department 12 causes the simulation log storage department 15 to accumulate the action results of executing the agent simulation by the agent simulation execution department 11 in step S11. Next, in step S13, the agent model generation department 12 learns the transition model, i.e., the agent model 16M, which is the internal state of each process, based on the action results accumulated in the simulation log storage department 15 and their execution order. Next, in step S14, the agent model generation department 12 saves the agent model 16M learned in step S13 in the agent model storage department 16.

[0082] (Risk analysis process)

[0083] Figure 8 It is a flowchart showing an example of the risk analysis process in the operation stage. In the operation stage, while actually controlling the operation equipment 40 according to the operation schedule to execute each process, risk analysis of problems with the operation schedule and the output of reports are carried out. The risk analysis process is frequently executed by the prediction department 13 and the risk assessment department 14 of the risk assessment system 10 at a predetermined cycle (e.g., once every few minutes), and the results are sent to the manager's terminal 18.

[0084] In actual operation in a factory or warehouse, etc., the tasks required for the equipment in one day are divided into dozens of batch units for processing. It is assumed that there are both definite elements with predetermined operation contents and indefinite elements defined by predicted values whose number and contents of the operation elements on the current day are not yet determined in this batch operation. Due to the existence of indefinite elements, even if an operation schedule based on the current assumed conditions is temporarily generated, it is necessary to frequently conduct risk analysis to grasp problem events and take countermeasures such as re-evaluating the operation schedule.

[0085] First, in step S21, the prediction unit 13 inputs the current situation (without assumed conditions). Next, in step S22, the prediction unit 13 sets 1 for the index n of the process. Next, in step S23, the prediction unit 13 activates the scheduler (not shown) of the planning system 2 to generate the best job schedule (start time, best resource allocation of the job equipment 40 and personnel, etc.) based on the job plan including predicted values under the current assumed conditions. Each job equipment 40 operates according to the job plan based on the information from the scheduler of the control system 1. And the current situation is fed back from the sensors provided in each job equipment 40. The control system 1 advances the process while correcting the start time and resource allocation, etc. of the job schedule based on the feedback information from the sensors.

[0086] Steps S24 to S31 are steps executed to predict the future progression of the job schedule generated in step S23. The prediction unit 13 receives the job schedule generated by the scheduler in step S23. The job schedule includes the allocation and order of batch units for each task, the job equipment numbers specifically allocated to each task, information recording the resource allocation timing, etc. Among these information, only the information used as parameters when learning the agent model 16M is used as the time series data An,k (k = 1, 2,...) of the task Tan input to the agent model 16M as the job result of the prediction process P(n). Figure 5 ) to use.

[0087] In step S24, the prediction unit 13 sets the initial conditions for the process P(n). Next, in step S25, the prediction unit 13 sets the assumed conditions for the process P(n). The assumed conditions refer to conditions such as problem events that cause a decrease in productivity, etc. in the process P(n).

[0088] Next, in step S26, the prediction unit 13 uses the data (tasks) of the process P(n) as the input to the agent model 16M. The input data of the process P(n) is the output data of the process P(n - 1), and based on the error distribution of the output of the process P(n - 1), a large number of input patterns are generated with a basic value + random error. Step S26 is repeated according to the number of input patterns. The input data (tasks) of the process P(1) is set to the given initial value.

[0089] Next, in step S27, the prediction unit 13 executes a surrogate calculation simulation using the surrogate model 16M a predetermined number of trial times under the current assumed conditions, and generates a plurality of output examples. The predetermined number of trial times is the same as the number of input patterns generated in step S26. The output includes the job end time, information related to productivity, problem event occurrence information, etc. Step S27 is executed multiple times for each input in step S26. The surrogate model 16M behaves probabilistically, outputs multiple patterns for each input, and generates a large number of output examples. That is, by executing the prediction multiple times using the surrogate model 16M for the same value, the fluctuations in the prediction results are reproduced.

[0090] Next, in step S28, the risk evaluation unit 14 calculates the occurrence probability of each problem event Qn,j of each process P(n) in the surrogate simulation in step S27. The occurrence probability of each problem event Qn,j is calculated based on the determination result of the occurrence of the problem event included in the output of the surrogate model 16M.

[0091] Next, in step S29, the prediction unit 13 obtains each output of the process P(n) in step S27, associates it with a label including the trial number for retrieval, and stores it in the simulation log storage unit 15. Each output of the process P(n) becomes the input of the process P(n + 1).

[0092] Next, in step S30, the prediction unit 13 sets the index n to +1. Next, in step S31, the prediction unit 13 determines whether n satisfies the end condition. In the present embodiment, the objects are processes P(1) to P(5) (n = 1 to 5). Therefore, when n = 6, the determination in step S31 is "yes", and the process proceeds to step S32. When n < 6, the process returns to step S24.

[0093] In step S32, the prediction unit 13 registers the simulation execution results of the loop of steps S24 to S31 together with the assumed conditions as an example in the simulation log storage unit 15. In step S32, for each loop of steps S22 to S34, the assumed conditions and the execution results are registered in correspondence. Thereby, it is possible to confirm over time how the execution results change in the process of sequentially incorporating the discovered problem events into the assumed conditions.

[0094] Next, in step S33, the risk evaluation unit 14 calculates the risk KPI (Key Performance Index) KPIn,j of each problem event Qn,j based on the occurrence probability pn,j of each problem event Qn,j of each process P(n) calculated in step S28 and registered as an example in the simulation log storage unit 15 in step S32, according to Equation (1). A, B, C, ka, and kb in Equation (1) are predetermined constants.

[0095] [Formula 1]

[0096]

[0097] Equation (1) is an example. As long as KPIn,j is an index that is larger when the occurrence probability pn,j is higher, larger when the remaining time t until the disposal plan action is shorter, and larger when the burden cost c at the time of generation is higher, it can also be an index based on other equations.

[0098] Next, in step S34, the risk evaluation unit 14 determines whether all the risk KPIs calculated in step S33 are below the threshold value Θ. When all the risk KPIs are below the threshold value Θ (step S34 "Yes"), the risk evaluation unit 14 transfers the process to step S35. When there is even one risk KPI greater than the threshold value Θ (step S34 "No"), the risk evaluation unit 14 transfers the process to step S36.

[0099] In step S35, the risk evaluation unit 14 performs a total processing of the simulation execution results registered as cases in the simulation log storage unit 15 in step S32, thereby performing a risk evaluation of the operation. Then, the risk evaluation unit 14 generates data for a report screen that presents the risk evaluation results to the manager and sends it to the terminal 18. In the total processing, assumed conditions, the occurrence probability of problem events, link information to the "response plan" in the operation schedule when a problem event occurs, the "response cost" when a problem event occurs, links to "other problem events affected" when a problem event occurs, etc. are targeted. Required information such as the "response plan", "response cost", and "other problem events affected" during the total calculation is stored in the problem event storage unit 17, for example, and is referred to during the total calculation. For a detailed example of the report screen, refer to Figure 9 、 Figure 10 and Figure 11 will be described later.

[0100] On the other hand, in step S36, the risk evaluation unit 14 adds the problem event corresponding to the risk KPI exceeding the threshold value Θ in step S34 to the assumed conditions and adds it to the recalculation candidate list. In the recalculation candidate list, for example, it includes the identification number of the problem event, the value of the risk KPI, execution time data, etc., and sorts the problem events in descending order of the risk KPI. The risk evaluation unit 14 sequentially extracts a predetermined number of problem events from the top of the recalculation candidate list (in the order of decreasing risk KPI) and adds them to the assumed conditions.

[0101] Then, in step S23, which is executed again, a work schedule is generated based on the assumptions that incorporate the problem event. In step S23, if a response plan has been defined for the problem event, the work schedule is regenerated based on the response plan. If no response plan has been defined, the work schedule is regenerated through scheduler optimization (resource reallocation, etc.). Then, steps S24 to S34 are executed to add cases.

[0102] In addition to executing step S35 on the condition that step S34 is "Yes", step S35 may be executed on the condition that the calculation end time is used.

[0103] Furthermore, in step S26, the uncertainty of the input is represented by multiple modes, but this is not limited to this. The uncertainty of the input can also be represented by a probability distribution represented by a probability density function. Similarly, in step S29, the uncertainty of the output is represented by multiple modes, but this is not limited to this. The uncertainty of the output can also be represented by a probability distribution represented by a probability density function. In other words, the input and output of the agent model 16M can be any of multiple modes of values and probability distributions.

[0104] (Dashboard display of risk analysis results)

[0105] Figure 9 、 Figure 10 and Figure 11 An example of a terminal display of a dashboard showing risk analysis results. Figure 9 、 Figure 10 and Figure 11 In step S35 ( Figure 8 ) outputted in the risk assessment results report screen for each process P(n). Risks include, for example, the probability of a problem event occurring in the agent simulation, the damage cost if a problem event occurs, and other possible problems that may occur as a result of the problem event. The risk analysis results dashboard is implemented using an application or browser executed on terminal 18.

[0106] like Figure 9 As shown, a process display 182 and a report display 183 are displayed on a display screen 181 of the terminal 18. In the process display 182, all the processes (processes P(1) to P(5)) of the SCM process S(2) that is the target of risk analysis in this example are displayed together with the work sequence. Figure 1 In the process display 182, as a result of the risk analysis, an identification mark 1821 for notifying that "there is risk" is displayed in the process determined to be likely to cause a problem event (in Figure 9The middle is an asterisk). "Having a risk above a certain level" means, for example, that the value of the risk KPIn,j is above the threshold Θ.

[0107] When the recognition mark 1821 is clicked, a report display 183 corresponding to problem events of each type is expanded and displayed. The report display 183 has an SCM impact display button 1831 and a device detail display button 1832.

[0108] When the SCM impact display button 1831 is clicked, a report display 1833 of problem events predicted to affect the SCM process S(2) displayed in the process display 182 is displayed. In addition, a countermeasure confirmation button 1834 and a report transmission function display 1835 are displayed in conjunction with the report display 1833. The problem events displayed here are, for example, those in step S33 ( Figure 8 ) where the risk KPI calculated is a predetermined number from the top.

[0109] In the report display 1833, "palletizing stagnation delay" is cited as a problem event that may occur in the future. During the process of performing the surrogate simulation execution ( Figure 8 step S27) 2500 times, it is "occurrence times 4 times" (occurrence probability is 4 / 2500). The "impact" of "palletizing stagnation delay" in process P(3) is that process P(3) is delayed by "average delay 32 seconds". As collateral risks affecting the subsequent processes P(4) and P(5), there are risks of "buffer congestion" in process P(4) and "departure time delay" in process P(5). The value of "average delay 32 seconds" is the average of the delay times simulated in each of the "occurrence times 4 times" trials.

[0110] In addition, as the "impact" of "palletizing stagnation delay" in process P(3), by adding the "average delay 32 seconds" of process P(3) to the assumed conditions (step S36 ( Figure 8 )) and re - executing the processes of steps S22 - S34 ( Figure 8 ), "buffer congestion" in process P(4) is predicted as a further "impact". And by adding the "buffer congestion" in process P(4) to the assumed conditions and re - executing the processes of steps S22 - S34, "departure time delay" in process P(5) is predicted as a further "impact". In this way, by adding problem events to the assumed conditions and executing steps S22 - S34, problem events that spread in a cross - process chain reaction can be predicted.

[0111] Although not shown in the figure, when the device detail display button 1832 is clicked, the work equipment (process P(3)) 40 - 3 that constitutes process P(3) where the recognition mark 1821 is displayed is displayed together with the operation sequence ( Figure 6Layout display of equipment and personnel.

[0112] When the countermeasure confirmation button 1834 is clicked, as Figure 10 shown, the countermeasure display 18341 is displayed. The countermeasure display 18341 represents a countermeasure for avoiding the problem event shown in the report display 1833. The countermeasure is stored as information corresponding to the problem event in the problem event storage unit 17. For the detailed display of the countermeasure, refer to Figure 10 the description later.

[0113] The "implementable time" displayed in the countermeasure confirmation button 1834 is the execution deadline by which the problem event can be prevented by implementing the countermeasure. The "impact" displayed in the countermeasure confirmation button 1834 indicates the magnitude of the risk KPI when the countermeasure is adopted.

[0114] When no countermeasure is defined for the problem event, the countermeasure confirmation button 1834 is not displayed.

[0115] The report transmission function display 1835 receives an instruction to start the following function, which is a function of designating a person in charge to send the report of the risk analysis result being displayed in the report display 183 together with a message. The user of the terminal 18 confirms the countermeasure and, if judged necessary, executes communication with relevant personnel.

[0116] When the countermeasure confirmation button 1834 ( Figure 9 ) is clicked, as Figure 10 shown, the countermeasure display 18341, the detailed display 18342, and the associated problem event confirmation button 18343 are displayed.

[0117] In the countermeasure display 18341, "palletizing stagnation delay" is cited as a problem event that may occur in the future, and as countermeasures, "(equipment) maintenance" and "(operation schedule) rescheduling" are cited. It is shown that the implementable time for these countermeasures is 12:35, the damage cost when the problem event occurs is 13,800, and the delay estimate when the problem event occurs is "30 p / 1 hour × 0.35 hours" for the palletizing stagnation delay. These information are, for example, information obtained by aggregating the action results of the agent simulation based on various information stored in the problem event storage unit 17.

[0118] The specific content of the countermeasure shown in the countermeasure display 18341 is displayed in the detailed display 18342, showing the responder, content, impact, etc. These information are stored, for example, in the problem event storage unit 17. According to the detailed display 18342, the countermeasure content of "Tarou Hitachi" converting "Robot AXX-VV" from a "palletizer" to a "depalletizer" is shown. In addition, since this conversion "reduces the palletizing productivity of process P(5)", "a correction that requires a change in the operation schedule of process P(4) and subsequent processes" is cited. These information are, for example, information obtained by aggregating the action results of the agent simulation based on various information stored in the problem event storage unit 17.

[0119] When the associated problem event confirmation button 18343 is clicked, as Figure 11 shown, the problem events derived from the problem event shown in the countermeasure display 18341 (i.e., in this example, the associated risks "Process P(4): Buffer congestion", "Process P(5): Departure time delay" shown in the report display 1833) Figure 9 ) are displayed in detail.

[0120] In addition, by clicking the associated problem event confirmation button 18343, the identification mark 1822 is displayed corresponding to the problem event of "Process P(4): Buffer congestion", and the identification mark 1823 is displayed corresponding to the problem event of "Process P(5): Departure time delay".

[0121] Figure 11 The report display 1836 showing the assumed associated problem events of process P(4) and the report display 1838 showing the assumed associated problem events of process P(5) that are displayed by clicking the associated problem event confirmation button 18343 are shown.

[0122] In the report display 1836, "operation buffer congestion" is cited as an associated problem event of process P(4) that may occur in the future in association with the "palletizing stagnation delay" of process P(3), and it is "the number of occurrences: 4 times" (occurrence probability 4 / 1500) during the execution of the agent simulation ( Figure 8 step S27) for 1500 times. In the report display 1838, "departure time delay" is cited as an associated problem event of process P(5) that may occur in association with the "palletizing stagnation delay" of process P(3) and the "operation buffer congestion" of process P(4), and it is "the number of occurrences: 2 times" (occurrence probability 2 / 1500) during the execution of the agent simulation ( Figure 8 step S27) for 1500 times. Other information can also be displayed in the report displays 1836 and 1838, but the illustration is omitted.

[0123] In the report transmission function display 1837, the designated recipient receives an instruction to send the report of the risk analysis result being displayed in the report display 1836 together with a message. The same function is also set in the report display 1838, but the illustration is omitted.

[0124] (Effect of the Embodiment)

[0125] In the present embodiment, a specific internal state that causes a decrease in job productivity found in the agent simulation is defined as a problem event, and an agent calculation model of the agent simulation is constructed so as to be able to reproduce the problem event. Then, an input value with fluctuations (uncertainty) is input to the agent model of the process, and an output value with fluctuations (uncertainty) due to the uncertainty of the input value and the probabilistic behavior of the agent model is used as the input value of the agent model of the next process. Therefore, it is possible to quickly simulate the future trends of jobs and the occurrence of problem events for a large number of patterns, and thus it is possible to quickly predict and evaluate the job risks including events with uncertainty.

[0126] In addition, it is possible to visualize the deviation of job progress from the job schedule and the prediction results of problem occurrences during the job, and it is possible to intuitively present to the user in an understandable manner the presence or absence and influence of alternative plans such as rescheduling of the job schedule, etc., and it is possible to help make a quick and accurate decision regarding the occurrence of problems.

[0127] (Hardware of the Computer 1000)

[0128] Figure 12 It is a hardware diagram showing a structural example of the computer 1000. For example, the computer 1000 implements a model generation device including an agent simulation execution unit 11 and an agent model generation unit 12, a job risk evaluation system including a prediction unit 13 and a risk evaluation unit 14, a terminal 18, or each device appropriately combined thereof.

[0129] The computer 1000 is a computer having a processor 1001 led by a CPU, a main storage device 1002, an auxiliary storage device 1003, a network interface 1004, an input device 1005, and an output device 1006 that are interconnected via an internal communication line 1009 such as a bus.

[0130] The processor 1001 is responsible for controlling the operation of the entire computer 1000. In addition, the main storage device 1002 is composed of a volatile semiconductor memory, for example, and is used as a working memory for the processor 1001. The auxiliary storage device 1003 is composed of a large-capacity non-volatile storage device such as a hard disk device, an SSD (Solid State Drive), or a flash memory, and is used to store various programs and data for a long time.

[0131] The executable program 1100 stored in the auxiliary storage device 1003 is loaded into the main storage device 1002 when the computer 1000 is started or when needed, and the processor 1001 executes the executable program 1100 loaded into the main storage device 1002, thereby implementing the above-described device for executing various processes.

[0132] The executable program 1100 can also be recorded on a non-transitory recording medium, read out from the non-transitory recording medium by a medium reading device, and loaded into the main storage device 1002. Alternatively, the executable program 1100 can also be obtained from an external computer via a network and loaded into the main storage device 1002.

[0133] The network interface 1004 is an interface device for connecting the computer 1000 to each network within the system or communicating with other computers. The network interface 1004 is constituted by, for example, a NIC (Network Interface Card) such as a wired LAN (Local Area Network) or a wireless LAN.

[0134] The input device 1005 is constituted by an indicating device such as a keyboard or a mouse, etc., and is used for a user to input various instructions or information to the computer 1000. The output device 1006 is constituted by, for example, a display device such as a liquid crystal display or an organic EL (Electro Luminescence) display, and a sound output device such as a speaker, etc., and is used for prompting the required information to the user when needed.

[0135] The present invention is not limited to the above-described embodiments, and includes various modification examples. For example, the above-described embodiments are embodiments described in detail for easily understanding the present invention, and are not limited to having all the structures described. In addition, as long as there is no contradiction, a part of the structure of a certain embodiment can be replaced with the structure of other embodiments, and the structure of other embodiments can be added to the structure of a certain embodiment. In addition, for a part of the structure of each embodiment, addition, deletion, replacement, combination, or dispersion of the structure can be performed. In addition, the structures and processes shown in the embodiments can be appropriately dispersed, combined, or replaced based on processing efficiency or installation efficiency.

[0136] Explanation of reference numerals

[0137] 1: Control system, 2: Planning system, 3: Control log storage unit, 10: Risk assessment system, 11: Agent simulation execution unit, 12: Agent model generation unit, 13: Prediction unit, 14: Risk assessment unit, 15: Simulation log storage unit, 16: Agent model storage unit, 17: Problem event storage unit, 18: Terminal.

Claims

1. An operation risk assessment system that performs risk assessment of an operation composed of processes, characterized in that: The operation risk assessment system has: A model storage unit that stores an agent model, which takes a first feature quantity of the operation as input and a second feature quantity of the operation calculated through a predetermined simulation of the process as output, and the agent model is a learned model of the relationship between the input and the output, and it agents the predetermined simulation with uncertainty, where the uncertainty means that for each input, the output corresponding to the same value of the input is different; A prediction unit that repeatedly performs a process of taking the first feature quantity as input and obtaining the second feature quantity as output from the agent model for a predetermined number of trial executions for the same value of the first feature quantity, thereby predicting a plurality of the second feature quantities with uncertainty for the same value of the first feature quantity in the process; and A risk assessment unit that performs risk assessment of the operation in the process based on a plurality of the second feature quantities with uncertainty predicted by the prediction unit, The operation includes a plurality of the processes, The agent model agents the predetermined simulation for each of the processes, The prediction unit inputs a plurality of the first feature quantities with uncertainty to the agent model of the process to predict a plurality of the second feature quantities with uncertainty in the process, and inputs the predicted plurality of the second feature quantities with uncertainty in the process as a plurality of the first feature quantities with uncertainty to the agent model of the next process to predict a plurality of the second feature quantities with uncertainty in the next process.

2. The operation risk assessment system according to claim 1, characterized in that: The predetermined simulation is an agent simulation, The agent model is a Bayesian neural network.

3. The operation risk assessment system according to claim 1, characterized in that: The uncertainty of a plurality of the first feature quantities and the uncertainty of a plurality of the second feature quantities are represented by a probability distribution.

4. The operation risk assessment system according to claim 1, characterized in that: The uncertainty of a plurality of the first feature quantities and the uncertainty of a plurality of the second feature quantities are represented by a set of a plurality of mode values.

5. The operation risk assessment system according to claim 1, characterized in that: The agent model takes the first feature quantity as input and the migration of the internal state of the process calculated through the predetermined simulation as output, and further learns the relationship between the input and the output, The prediction unit uses the agent model to predict a plurality of the second feature quantities with uncertainty and the migration of the internal state in the process together in an operation schedule generated under predetermined assumption conditions, The risk assessment unit determines whether the internal state predicted by the prediction unit to migrate is equivalent to a problem event predefined as a specific internal state that reduces the productivity of the process.

6. The operation risk assessment system according to claim 5, characterized in that: The risk evaluation unit calculates the occurrence probability of the problem event, and this occurrence probability is used to evaluate the risk of the problem event. When a predetermined index based on this occurrence probability exceeds a threshold value, the risk evaluation unit notifies the scheduler to regenerate the job schedule after adding the problem event to the predetermined assumed conditions. The prediction unit re-predicts together the migrations of the plurality of second feature amounts including uncertainties and the internal state in the process based on the predetermined assumed conditions to which the problem event is added.

7. The job risk evaluation system according to claim 6, wherein when the internal state is equivalent to the problem event, the risk evaluation unit generates data for notifying at least any one of the name of the problem event, the occurrence frequency of the problem event, the predetermined number of trials, the occurrence probability of the problem event, other problem events that occur in chain with the problem event, and the countermeasure plan for the problem event to the user, and sends it to the terminal of the user. The terminal displays a screen based on the received data.

8. A model generation device that generates a prediction model for predicting feature amounts of a job composed of processes, wherein the model generation device includes: a simulation execution unit that executes a predetermined simulation of the process that takes the first feature amount of the job as an input and outputs the migration of the internal state of the process and the second feature amount of the job; and a surrogate model generation unit that generates a surrogate model as the prediction model, and the surrogate model is a learned model of the relationship between the input and the output, and surrogates the predetermined simulation having uncertainties, where the uncertainties mean that for each input, the outputs corresponding to the inputs of the same value of the first feature amount are different. Through the prediction model, when the first feature amount is used as an input and the internal state of the process migrates to a problem event that is predefined as a specific internal state that reduces the productivity of the process, the discovery of the problem event in the process is measured.

9. A job risk evaluation method performed by a job risk evaluation system that evaluates the risk of a job composed of processes, wherein the job risk evaluation system has a model storage unit that stores a surrogate model that takes the first feature amount of the job as an input and outputs the second feature amount of the job calculated by the predetermined simulation of the process, and the surrogate model is a learned model of the relationship between the input and the output, and surrogates the predetermined simulation having uncertainties, where the uncertainties mean that for each input, the outputs corresponding to the inputs of the same value are different. The job risk evaluation method includes the following steps: a prediction step of repeatedly performing a process of taking the second feature amount as an output from the surrogate model with the first feature amount as an input for a predetermined number of trials for the first feature amount of the same value, thereby predicting a plurality of the second feature amounts having uncertainties with respect to the first feature amount of the same value in the process. A risk assessment step that performs a risk assessment of the operation in the process based on a plurality of the second characteristic quantities with uncertainty predicted by the prediction step. The operation includes a plurality of the processes. The surrogate model surrogates the predetermined simulation for each of the processes. In the prediction step, a plurality of the first characteristic quantities with uncertainty are input to the surrogate model of the process to predict a plurality of the second characteristic quantities with uncertainty in the process, and the predicted plurality of the second characteristic quantities with uncertainty in the process are input as a plurality of the first characteristic quantities with uncertainty to the surrogate model of the next process to predict a plurality of the second characteristic quantities with uncertainty in the next process.

10. The operation risk assessment method according to claim 9, wherein The predetermined simulation is a surrogate simulation. The surrogate model is a Bayesian neural network.

11. The operation risk assessment method according to claim 9, wherein The uncertainty of the plurality of the first characteristic quantities and the uncertainty of the plurality of the second characteristic quantities are represented by probability distributions.

12. The operation risk assessment method according to claim 9, wherein The uncertainty of the plurality of the first characteristic quantities and the uncertainty of the plurality of the second characteristic quantities are represented by a set of a plurality of mode values.

13. The operation risk assessment method according to claim 9, wherein The surrogate model takes the first characteristic quantity as an input, takes the transition of the internal state of the process calculated by the predetermined simulation as an output, and further learns the relationship between the input and the output. In the prediction step, in the operation schedule generated under predetermined assumption conditions, the surrogate model is used to predict a plurality of the second characteristic quantities with uncertainty in the process and the transition of the internal state together. In the risk assessment step, it is determined whether the internal state predicted by the prediction step to undergo a transition corresponds to a problem event predefined as a specific internal state that reduces the productivity of the process.

14. The operation risk assessment method according to claim 13, wherein In the risk assessment step, the occurrence probability of the problem event is calculated, and this occurrence probability is used to evaluate the risk of the problem event. When a predetermined index based on this occurrence probability exceeds a threshold value, the scheduler is notified to regenerate the operation schedule after adding the problem event to the predetermined assumption conditions. In the prediction step, based on the predetermined assumption conditions to which the problem event is added, a re - prediction is performed on a plurality of the second characteristic quantities with uncertainty in the process and the transition of the internal state together.

15. The operation risk assessment method according to claim 14, wherein In the risk assessment step, when the internal state is equivalent to the problem event, data for notifying at least any one of the name of the problem event, the occurrence frequency of the problem event, the predetermined number of trial runs, the occurrence probability of the problem event, other problem events that are chained to the problem event, and the countermeasure plan for the problem event to the user is generated and sent to the terminal of the user. The terminal displays a screen based on the received data.

16. A job risk assessment program product, characterized in that the job risk assessment program product is used to cause a computer to function as the job risk assessment system according to any one of claims 1 to 7.

Images