Orchestration system, computer, orchestration program, and orchestration method

Abstract

To improve inspection accuracy by optimizing parameters of an inspection module.SOLUTION: An orchestration system comprising a processing unit and a storage unit, wherein the processing unit acquires, from a plurality of inspection modules arranged in a production line in a predetermined arrangement order, inspection result data on an inspection object arranged in the production line, and inputs the acquired inspection result data to a learning model associated with a first inspection module, determine which of a plurality of clusters the inspection object belongs to based on an output from the learning model, and acquire a setting value of a parameter used by an inspection module arranged in the production line other than the first inspection module, the setting value being stored in the storage unit in accordance with the cluster to which the inspection object belongs.SELECTED DRAWING: Figure 7

Description

[Technical Field]

[0001] The present invention relates to an orchestration system, a computer, an orchestration program, and an orchestration method. [Background technology]

[0002] Patent Document 1 describes a production line management method in which various processing devices incorporated in each processing step of a production line that produces products by packing subdivided items into bags record the processing results of each processed item in association with each item, and identify problematic processing devices from the processing results obtained for each item at each processing device. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 9-301327 Summary of the Invention [Problem to be solved by the invention]

[0004] In some cases, multiple inspection modules are arranged on a production line, etc., to inspect objects flowing along the production line from multiple viewpoints. In managing the multiple inspection modules, if the difficulty in judgment in one inspection module can be compensated for by adjusting the settings of another inspection module, the accuracy of judgment as a whole can be improved.

[0005] The present invention has been made in view of the above circumstances, and provides an orchestration system, a computer, an orchestration program, and an orchestration method that can improve inspection accuracy by optimizing parameters of an inspection module. [Means for solving the problem]

[0006] In order to achieve the above-mentioned object, the orchestration system, computer, orchestration program, and orchestration method of the present invention are characterized by the following [1] to [8].

[0007] [1] An orchestration system (100) having a processing unit and a storage unit, The processing unit Obtaining inspection result data for the objects to be inspected that are arranged on the production line from a plurality of inspection modules (50, 60, 70) that are arranged on the production line in a predetermined order; inputting the acquired inspection result data into a learning model associated with a first inspection module, and determining to which of a plurality of clusters the inspection object belongs based on an output from the learning model; acquiring setting values ​​of parameters used by inspection modules arranged in the production line other than the first inspection module, the setting values ​​being stored in the storage unit according to the cluster to which the inspection object belongs; Orchestration system.

[0008] [2] The plurality of inspection modules include at least one of a metal detector (50), an X-ray inspection machine (60), and a weight checker (70). [1] The orchestration system described in.

[0009] [3] The inspection modules arranged in the production line other than the first inspection module are inspection modules arranged downstream of the first inspection module in the production line. [1] The orchestration system described in.

[0010] [4] The inspection modules arranged in the production line other than the first inspection module include an inspection module arranged next to the first inspection module in the production line. [1] The orchestration system described in.

[0011] [5] The processing unit further transmits the set values ​​of the parameters to inspection modules arranged in the production line other than the first inspection module. An orchestration system according to any one of [1] to [4].

[0012] [6] A computer (2) having a processor and memory, The processor: Obtaining inspection result data for the objects to be inspected that are arranged on the production line from a plurality of inspection modules (50, 60, 70) that are arranged on the production line in a predetermined order; inputting the acquired inspection result data into a learning model associated with a first inspection module, and determining to which of a plurality of clusters the inspection object belongs based on an output from the learning model; acquiring setting values ​​of parameters used by inspection modules arranged in the production line other than the first inspection module, the setting values ​​being stored in the memory according to the cluster to which the object to be inspected belongs; computer.

[0013] [7] 1. An orchestration program comprising: A computer (2) having a processor and a memory, a function of acquiring inspection result data for the objects to be inspected that are arranged on the production line from a plurality of inspection modules (50, 60, 70) that are arranged on the production line in a predetermined order; a function of inputting the acquired inspection result data into a learning model associated with a first inspection module, and determining to which of a plurality of clusters the inspection object belongs based on the output from the learning model; a function of acquiring setting values ​​of parameters used by inspection modules arranged in the production line other than the first inspection module, the setting values ​​being stored in the memory according to the cluster to which the inspection object belongs; An orchestration program that makes this possible.

[0014] [8] An orchestration method using a computer (2) having a processor and a memory, comprising: the processor: Obtaining inspection result data for the objects to be inspected that are arranged on the production line from a plurality of inspection modules (50, 60, 70) that are arranged on the production line in a predetermined order; inputting the acquired inspection result data into a learning model associated with a first inspection module, and determining to which of a plurality of clusters the inspection object belongs based on an output from the learning model; acquiring setting values ​​of parameters used by inspection modules arranged in the production line other than the first inspection module, the setting values ​​being stored in the memory according to the cluster to which the object to be inspected belongs; Orchestration method. [Effects of the Invention]

[0015] According to the present invention, the inspection accuracy can be improved by optimizing the parameters of the inspection module. [Brief explanation of the drawings]

[0016] [Figure 1] FIG. 1 is a conceptual diagram illustrating an inspection module according to an embodiment of the present disclosure. [Figure 2] FIG. 10 illustrates inputs, outputs, and parameters in a detection process according to one embodiment of the present disclosure. [Figure 3] FIG. 1 is a conceptual diagram illustrating the orchestration of multiple inspection modules according to an embodiment of the present disclosure. [Figure 4]FIG. 1 is a conceptual diagram illustrating how input data is handled in hyperparameter optimization according to an embodiment of the present disclosure. [Figure 5] FIG. 1 is a conceptual diagram illustrating an orchestration system according to an embodiment of the present disclosure. [Figure 6] FIG. 1 is a conceptual diagram illustrating group-wise optimization according to an embodiment of the present disclosure. [Figure 7] FIG. 1 is a conceptual diagram illustrating an orchestration system according to an embodiment of the present disclosure. [Figure 8] FIG. 10 is a diagram illustrating setting values ​​indicating optimal settings for a downstream inspection module according to an embodiment of the present disclosure. [Figure 9] FIG. 10 is a diagram illustrating a mathematical model in accordance with an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0017] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0018] FIG. 1 is a conceptual diagram illustrating an inspection module according to one embodiment of the present disclosure.

[0019] A production line is provided with a plurality of inspection modules for inspecting objects to be inspected. In FIG. 1, 50, 60, and 70 each correspond to an inspection module. 50 is a metal detector. 60 is an X-ray inspection machine. 70 is a checkweigher. The inspection modules provided on the production line are not limited to 50, 60, and 70. FIG. 1 illustrates an example of a multi-function machine that integrates a plurality of inspection modules into a single device, but this is not limiting. A plurality of inspection modules may also be provided on the production line as separate devices. In this specification, the inspection module included in the multifunction peripheral and the inspection module arranged as a separate device are collectively referred to as the inspection module.

[0020] The inspection module may judge whether the inspected item is OK or NG. For example, if the inspected item is good, it is OK, and if it is not good, it is NG. It may also detect foreign objects (inspected items) that have been mixed into the production line as NG.

[0021] When determining whether an object is OK or NG using a single inspection module, the accuracy of the determination can be improved by using another inspection module in combination. For example, metal detector 50 has a weak metal reaction to small objects, making it difficult to detect them. Lowering the threshold for NG determination makes it possible to detect small objects as NG, but it also increases the possibility of determining that a good object is NG. Setting the optimal threshold is not an easy task.

[0022] Here, the X-ray inspection machine 60 is suited to detecting small objects to be inspected, so if the OK / NG judgment result from the metal detector 50 and the OK / NG judgment result from the X-ray inspection machine 60 are used together to make the judgment, the overall judgment accuracy will improve.

[0023] For example, if the objects to be inspected are scallops and it is desired to classify medium-sized scallops as M class and large scallops as L class, it is difficult to distinguish between M and L class using the X-ray inspection machine 60. However, if a weight sorter 70 is used in addition, the weight sorter 70 can distinguish between M and L class by weight, thereby improving the overall accuracy of the judgment.

[0024] If the object to be inspected is a metal ball (foreign object), it is difficult to detect the metal ball using the image processing of the X-ray inspection machine 60, but by adjusting the AC magnetic field frequency of the metal detector 50, the metal ball can be detected, thereby improving the overall accuracy of the judgment.

[0025] The above is merely an example, but by using a plurality of inspection modules in combination rather than a single inspection module, it is possible to improve the accuracy of determining the object to be inspected.

[0026] However, when a plurality of inspection modules are used, the ease of detection differs depending on the object to be inspected, which raises another problem of how to set the parameters for each inspection module.

[0027] Therefore, the orchestration system of the present disclosure enables the parameters of multiple inspection modules to be automatically set to appropriate values. This process of setting parameters appropriately is referred to as "optimization" in this specification. Optimization includes parameter optimization and hyperparameter optimization.

[0028] (parameter) The parameters are, for example, as follows, but are not limited to these: -Coefficients and fixed values ​​of the model used by the inspection module A number specifying the contribution of the input How much bias to add Parameters are optimized by training (learning) the model corresponding to the inspection module. For example, the kernel coefficients of the CNN (Convolutional Neural Network) corresponding to the inspection module are optimized. Parameter optimization is often achieved by learning while changing parameters to minimize the loss function.

[0029] (hyperparameters) Examples of hyperparameters include, but are not limited to, the following: -Parameters that change the model calculation formula -Parameters that define numerical limits The hyperparameters are optimized by searching for the best fit. For example, the architecture activation function of a CNN (Convolutional Neural Network) is optimized. Hyperparameter optimization is often achieved by maximizing an evaluation function.

[0030] To properly set parameters, "hyperparameter optimization" in artificial intelligence (AI), particularly machine learning (ML), is used. There are many different methods for hyperparameter optimization (HPO). Therefore, those skilled in the art can select an appropriate optimization method depending on the actual application. Generally, the following are required for hyperparameter optimization: Input data Mathematical Model Evaluation function (index) and objective The range of possible values ​​for the hyperparameters

[0031] Here is a specific example of hyperparameter optimization. For example, when optimizing an image processing algorithm, the following information is given to the system as necessary for optimization: Hyperparameters: filter number, kernel size Input data: PNG, JPG, etc. · Mathematical model: Image processing filter Evaluation function and objective: An evaluation function that outputs the difference between the minimum detection limit at which a foreign object can be detected and the maximum detection limit at which a foreign object cannot be detected. A filter that maximizes this difference is searched for. Range of possible values ​​for hyperparameters: (list of filter numbers)

[0032] (Evaluation functions, indicators, objectives) The optimization objective is determined according to the production requirements of the production line. For example, the detection rate and the false detection rate are used as the production requirements. Then, the evaluation function is determined based on the detection rate and the false detection rate.

[0033] Evaluation function: A function to calculate the TP rate (detection rate) and FP rate (false detection rate) Evaluation indicators: TP rate and FP rate Optimization objective: maximize TP rate and minimize FP rate.

[0034] In order to calculate the TP rate and FP rate, actual OK / NG information is required, so in order to perform optimization based on the above evaluation function, training data with labels indicating OK or NG is required.

[0035] (Mathematical Model) To calculate the TP rate and FP rate, a model for determining whether a test is OK or NG is required, i.e., a mathematical model of the detection process performed by the detection module. For the three types of detection modules described above, the detection process performed by each detection module can be expressed as three steps: a measurement step, a processing step, and a judgment step. For example, in the measurement (M) step, the detection module passes the test object through a non-contact field such as a magnetic field or X-rays, and measures the field response with a sensor to obtain a signal. In the processing (P) step, the detection module processes the obtained signal and extracts the most important features. In the judgment (J) step, the detection module compares the extracted features with standard values ​​to determine whether the test object is normal or abnormal. The input data, output data, and parameters for each of the measurement step, processing step, and judgment step are, for example, as shown in Figure 2. Figure 2 illustrates the inputs, outputs, and parameters of the detection process according to one embodiment of the present disclosure.

[0036] FIG. 3 is a conceptual diagram illustrating the orchestration of multiple inspection modules according to one embodiment of the present disclosure.

[0037] Generally, orchestration means automatically managing the resources of multiple entities (programs, devices, agents, etc.) For example, in service orchestration, to prevent downtime of an online service, when a program goes down, a backup is quickly started, routing is corrected, and the necessary CPU resources are allocated.

[0038] This disclosure deals with an orchestration system that performs orchestration when multiple entities are test modules and automatic resource management is performed by adjusting parameter settings. The orchestration system of this disclosure optimizes multiple test modules together, rather than a single test module. In other words, the orchestration of this disclosure controls the horizontal collaboration of multiple test modules. In this respect, the orchestration of this disclosure differs from the process of automatically configuring a single test module.

[0039] 3, the orchestration system of the present disclosure acquires as input data signal data from the metal detector 50, raw transmission images from the X-ray inspection machine 60, and signal data from the checkweigher 70. To ensure horizontal coordination of data from the metal detector 50, X-ray inspection machine 60, and checkweigher 70 as an orchestration system, it is better to acquire data by using all inspection modules lined up on the production line together, rather than operating each detection module one by one. Figure 3 shows how data is acquired from all inspection modules.

[0040] The orchestration system of the present disclosure may further acquire information such as influence values ​​and phase values ​​from the metal detector 50, processed images from the X-ray inspection machine 60, weights from the weighing machine 70, OK / NG judgment results based on this information, and parameters used when the data was acquired. This information may be useful in narrowing down the search space. The orchestration system of the present disclosure may also acquire information other than these.

[0041] FIG. 4 is a conceptual diagram illustrating how input data is handled in hyperparameter optimization according to an embodiment of the present disclosure.

[0042] Considering that hyperparameter optimization is actually performed in a factory, it is difficult to obtain label data at times other than during operational checks. Therefore, semi-supervised learning can be performed using both operational checks and history supervised data and unsupervised data. In particular, transduction is preferable. Supervised learning, semi-supervised learning, and transduction are well known techniques to those skilled in the art, so detailed explanations will be omitted.

[0043] If a large amount of supervised data is available, the optimization process can be performed without semi-supervised learning.

[0044] FIG. 5 is a conceptual diagram illustrating an orchestration system according to an embodiment of the present disclosure.

[0045] The orchestration system 100 of the present disclosure includes an inspection unit 1, an orchestration unit 2, a display unit 3, and a server 4.

[0046] The inspection unit 1 includes a plurality of inspection modules arranged on the production line. In Fig. 5, the inspection unit 1 includes three inspection modules: a metal detector 50, an X-ray inspection machine 60, and a weight checker 70.

[0047] The orchestration unit 2 may be a computer including at least a processing unit and a storage unit. The processing unit included in the orchestration unit 2 loads and executes an orchestration program stored in the storage unit to realize various functions of the orchestration unit 2. The processing unit may include a processor. The storage unit may include a memory.

[0048] If the accumulation of supervised data is insufficient, the orchestration unit 2 performs semi-supervised learning such as the above-mentioned propagation to increase the amount of supervised data.

[0049] (1) The orchestration unit 2 selects one parameter candidate from the parameter search group. (2) The orchestration unit 2 sets a model based on the parameter candidates and inputs training data based on the test result data into the model. (3) The orchestration unit 2 evaluates the output results from the model to measure the appropriateness of the parameter candidates. (4) Repeat from (1) above, and select the next parameter candidate to try based on its appropriateness. As described above, the orchestration unit 2 repeats input, output, evaluation, and parameter search in the model to optimize the parameters for the inspection module. The optimized parameters may be stored in memory as optimal settings. A value indicating this optimal setting may be displayed on the display unit 3.

[0050] The display unit 3 is, for example, a touch panel and includes a display device and an input device. The user inputs information such as the detection rate into the display unit 3. Furthermore, based on instructions input by the user, the display unit 3 may transmit information indicating an evaluation request to the orchestration unit 2. Upon receiving the evaluation request, the orchestration unit 2 starts evaluation processing of the output data from the model. The display unit 3 displays the information received from the orchestration unit 2 on the display device. The information from the orchestration unit 2 may be, for example, the setting values ​​of the optimal settings, analysis data after the transmission processing, etc. The display unit 3 may transmit the setting values ​​of the optimal settings acquired from the orchestration unit 2 to the inspection unit 1. The inspection unit 1 optimizes the inspection module based on the setting values ​​of the optimal settings acquired from the display unit 3, i.e., sets the optimal setting values ​​of the parameters in the inspection module. Additionally, the display unit 3 may acquire the inspection results by the inspection unit 1, for example, information indicating OK / NG, from the inspection unit 1 and display it on the display device.

[0051] The display unit 3 may be a device other than a touch panel that can input and output information. For example, the display unit 3 may be a computer equipped with a processor, a memory, an input device, and a display device.

[0052] The server 4 may be located on-site or on the cloud. The server 4 includes a processor and a memory. The server 4 transmits supervised data to the orchestration unit 2. The orchestration unit 2 performs the transmission process described above using the received supervised data. The server 4 may also acquire and store various types of information from the orchestration unit 2. For example, the various types of information include information indicating the analysis results after the transmission process, information indicating the evaluation results after the evaluation process, and setting values ​​for optimal settings, but other information may also be included.

[0053] In the figure, the setting values ​​for the optimal settings are transmitted from the orchestration unit 2 to the inspection unit 1 via the display unit 3, but the setting values ​​for the optimal settings may also be transmitted directly from the orchestration unit 2 to the inspection unit 1. The setting values ​​for the optimal settings may include setting values ​​for multiple inspection modules. In other words, optimization is performed to ensure horizontal cooperation between multiple inspection modules. In other words, transmitting the setting values ​​for the optimal settings corresponds to orchestration.

[0054] (Group-wise optimization) Data corresponding to the objects to be inspected may be divided into groups, and optimal parameters may be searched for for each group. Figure 6 is a conceptual diagram illustrating optimization for each group according to an embodiment of the present disclosure.

[0055] Assume that inspection module #1, inspection module #2, and inspection module #3 are arranged side by side on a production line. The orchestration unit 2 optimizes parameters using the data acquired from inspection module #1 among the inspection result data acquired from the inspection unit 1 (St101). After optimization, the orchestration unit 2 performs clustering and classifies the inspection result data into multiple clusters (St102).

[0056] The orchestration unit 2 extracts clusters containing detection failures from the classified clusters (St103).

[0057] The orchestration unit 2 optimizes the parameters for each cluster using the data acquired from the inspection module #2 among the inspection result data acquired from the inspection unit 1. At this time, the orchestration unit 2 performs optimization using only the data that belong to the same cluster in the previous clustering result (St104).

[0058] As an example, suppose that the clusters with detection failures extracted in step St103 above are cluster A and cluster B. In this case, the orchestration unit 2 optimizes the parameters using only the data that belongs to cluster A and is acquired from inspection module #2 out of the inspection result data from the inspection unit 1. Also, the orchestration unit 2 optimizes the parameters using only the data that belongs to cluster B and is acquired from inspection module #2 out of the inspection result data from the inspection unit 1.

[0059] The orchestration unit 2 repeats the above process for each inspection module arranged in a line on the production line. For example, after the optimization in step St104, the orchestration unit 2 performs clustering to classify the inspection result data into multiple clusters. The orchestration unit 2 extracts clusters containing detection failures from the multiple classified clusters. The orchestration unit 2 optimizes parameters for each cluster using data acquired from inspection module #3 among the inspection result data from the inspection unit 1. At this time, the orchestration unit 2 performs optimization using only data that falls into the same cluster in the previous clustering results. Even if there are four or more inspection modules, the same procedure is used to sequentially classify the inspection result data for a certain inspection module into multiple clusters, and then optimization is performed using the inspection data for the subsequent inspection modules, separated by cluster.

[0060] FIG. 7 is a conceptual diagram illustrating an orchestration system according to an embodiment of the present disclosure.

[0061] Of the orchestration system 100 shown in FIG. 7, detailed explanations of the same parts as those in FIG. 5 will be omitted.

[0062] In FIG. 7, a series of optimization processes is performed for each cluster, evaluating the output results from the model and searching for parameters.

[0063] The orchestration unit 2 performs the following processes (1) to (4) for each model. (1) The orchestration unit 2 selects one parameter candidate from the parameter search group. (2) The orchestration unit 2 sets a model based on the parameter candidates and inputs training data based on the test result data into the model. (3) The orchestration unit 2 evaluates the output results from the model to measure the appropriateness of the parameter candidates. (4) Repeat from (1) above, and select the next parameter candidate to try based on its appropriateness.

[0064] The orchestration unit 2 further uses a model (a model associated with the preceding inspection module) to which supervised data for the preceding inspection module is input to determine to which of a plurality of clusters the object under test belongs, i.e., classifies the object under test into a plurality of clusters.

[0065] The orchestration unit 2 inputs the supervised data for the subsequent inspection module into the model associated with the subsequent inspection module ((2) above). This input is performed for each cluster after dividing the data by cluster.

[0066] The above process is performed for each inspection module arranged on the production line, and the setting values ​​indicating the optimal settings are stored in the memory of the orchestration unit 2. The setting values ​​indicating the optimal settings are transmitted from the orchestration unit 2 to the inspection unit 1 via, for example, the display unit 3. The inspection unit 1 sets the setting values ​​indicating the optimal settings in each inspection module.

[0067] FIG. 8 is a diagram illustrating setting values ​​indicating optimal settings for a subsequent inspection module according to an embodiment of the present disclosure. The subsequent inspection module is, for example, inspection module #2. The previous inspection module is, for example, inspection module #1. Assume that the objects to be inspected are classified into four clusters, clusters A to D, through clustering for the previous inspection module. Since optimization is performed for each classified cluster for the subsequent inspection module, setting values ​​indicating optimal settings are stored in memory for each cluster. As shown in the figure, the setting values ​​indicating optimal settings may include setting values ​​for multiple inspection modules arranged downstream of the previous inspection module in the production line.

[0068] (Inference stage) The above describes the learning phase of the orchestration system of the present disclosure. Next, the inference phase will be described.

[0069] The inspection objects are sent down the production line and inspected by a first inspection module arranged in association with the production line. The orchestration unit 2 acquires inspection result data for the inspection objects arranged on the production line from the first inspection module.

[0070] The orchestration unit 2 inputs the acquired inspection result data into a model associated with the first inspection module, and determines to which of a plurality of clusters the inspection object belongs based on the output from the model.

[0071] The orchestration unit 2 acquires the setting values ​​of the parameters stored in the memory according to the cluster to which the object to be inspected belongs, and used by the inspection modules arranged in the production line other than the first inspection module.

[0072] The orchestration unit 2 transmits the acquired setting values ​​of the parameters used by the inspection modules to the inspection modules arranged in the production line other than the first inspection module.

[0073] The setting values ​​for the inspection module may be set automatically, or the setting values ​​may be output to the user by displaying them on the display unit 3, and then the user may adjust the inspection module.

[0074] The inspection modules arranged in the production line other than the first inspection module may be inspection modules arranged downstream of the first inspection module in the production line. With reference to Fig. 7, for example, the first inspection module is a metal detector 50. The inspection modules arranged downstream of the first inspection module are an X-ray inspection machine 60 and a checkweigher 70.

[0075] The inspection modules arranged in the production line other than the first inspection module may include an inspection module arranged next to the first inspection module in the production line. With reference to FIG. 7 , the first inspection module may be a metal detector 50. In this case, the inspection module arranged next to the first inspection module is an X-ray inspection machine 60. The first inspection module may also be an X-ray inspection machine 60. In this case, the inspection module arranged next to the first inspection module is a checkweigher 70.

[0076] Note that the inspection modules arranged in the production line other than the first inspection module may be inspection modules arranged upstream of the first inspection module in the production line. Since the orchestration system 100 acquires inspection result data from multiple inspection modules, it may perform clustering of the objects to be inspected for the inspection modules arranged downstream, and then set parameters for the inspection modules arranged upstream according to the classified clusters.

[0077] (Supplementary explanation of the mathematical model) The model for determining whether a product passes or fails to calculate the TP rate and FP rate, i.e., the mathematical model of the detection process performed by the detection module, has been described above. As a supplement, we will explain the implementation of the mathematical model in a program and an example of the mathematical model for an X-ray inspection machine (60).

[0078] 1. Model Implementation As mentioned above, the process of each detection module includes three steps: measurement (M), processing (P), and judgment (J). In principle, each step is implemented by an independent function, with the output of the M function as the input to the P function, and the output of the P function as the input to the J function.

[0079] The output of the J function (OK / NG judgment) is passed to the evaluation function for parameter candidates. The parameters for each step of M / P / J are used as internal variables or function arguments, and are changed and used when performing the parameter optimization described above. If there are already determined parameters that do not need to be optimized, their values ​​can be fixed. When implementing this function using object-oriented programming, the M / P / J functions for one model and related parameters can be made into a class.

[0080] 1.1 Measurement (M) Steps and Functions Unlike the processing (P) and judgment (J) steps, the original input of the measurement (M) step is not data but a physical object, such as an inspected product. Therefore, the measurement (M) step cannot be modeled as a typical function that receives numerical data as input, processes it, and outputs the processed results. Rather, the measurement (M) function is a function that generates new data according to parameters. For example, in the case of a model of an X-ray inspection machine 60, it generates an image with contrast according to the parameter of X-ray energy. Data generation can be performed using technologies such as generative AI. However, data can also be generated in a simpler way, such as by processing existing data acquired under certain parameters to convert it into equivalent data acquired under different parameters.

[0081] It is also possible to completely remove the measurement (M) step from parameter optimization. In this case, measurement data, which is the output of the measurement (M) step, can be acquired by an actual device included in the orchestration system 100, and the measurement data can be read when optimizing the processing (P) step or the decision (J) step.

[0082] Three different embodiments of the measurement (M) step are illustrated in Figure 9. Figure 9 provides additional explanatory diagrams of a mathematical model according to one embodiment of the present disclosure.

[0083] 9 shows three types of embodiments: the generating function M model method, the on-hand data conversion M method, and the external M step method. The generating function M model method shows an embodiment in which the measurement (M) function is used as a function to generate new data. The on-hand data conversion M model method shows an embodiment in which on-hand data acquired with certain parameters is converted by some processing to produce equivalent data that appears to have been acquired with other parameters. The external M step method shows an embodiment in which the measurement (M) step is removed from the parameter optimization.

[0084] 2. Example of an X-ray inspection machine model 2.1 Measurement (M) step M function We use the "M model method for handheld data conversion" shown in Figure 9. The input is a transmission image acquired with energy E, and the arguments and parameters are energy F. Using any processing, the input image is converted into an equivalent transmission image taken with F and output.

[0085] 2.2 P-function of the processing (P) step It takes a transparent image input and applies a feature extraction image processing filter. The arguments and parameters of the function are the type of filter and its parameters. The parameters mean, for example, the kernel size. The filter can be a single-stage or multi-stage filter. More than one filter can be used. The output is a grayscale image showing abnormalities in the image.

[0086] 2.3 J function of the decision (J) step The same number of limit parameters are used as arguments and parameters for the J function according to the number of input grayscale images. After input, the maximum brightness value of each image is compared with the equivalent limit, and if it exceeds the equivalent limit, an "NG" judgment is output. If the maximum brightness value of each image does not exceed the equivalent limit, an "OK" judgment is output. Note that it is also possible to output as a Boolean value such as OK = 0 and NG = 1. If there are multiple images, it is also possible to output an AND, OR, etc. of all comparison results.

[0087] Although various embodiments have been described above with reference to the drawings, it goes without saying that the present disclosure is not limited to such examples. It is clear to those skilled in the art that various modifications and alterations can be made within the scope of the claims, and it is understood that these also fall within the technical scope of the present disclosure. For example, the steps in the methods disclosed herein may be performed in any order as long as no contradictions arise. Furthermore, the components in the above embodiments may be combined in any order as long as they do not deviate from the spirit of the disclosure. [Explanation of symbols]

[0088] 1. Inspection Department 2 Orchestration Department 3 Display section 4 Server 50 Metal Detector 60 X-ray inspection machine 70 Weight checker 100 Orchestration Systems

Claims

1. An orchestration system (100) having a processing unit and a storage unit, The processing unit Obtaining inspection result data for the objects to be inspected that are arranged on the production line from a plurality of inspection modules (50, 60, 70) that are arranged on the production line in a predetermined order; inputting the acquired inspection result data into a learning model associated with a first inspection module, and determining to which of a plurality of clusters the inspection object belongs based on an output from the learning model; acquiring setting values ​​of parameters used by inspection modules other than the first inspection module, the setting values ​​being stored in the storage unit according to the cluster to which the inspection object belongs; Orchestration system.

2. The plurality of inspection modules include at least one of a metal detector (50), an X-ray inspection machine (60), and a weight checker (70). The orchestration system of claim 1 .

3. the inspection modules other than the first inspection module are inspection modules arranged downstream of the first inspection module in the production line. The orchestration system of claim 1 .

4. the inspection modules arranged in the production line other than the first inspection module include an inspection module arranged next to the first inspection module in the production line; The orchestration system of claim 1 .

5. The processing unit further transmits the set value of the parameter to an inspection module other than the first inspection module that is arranged in the production line. The orchestration system according to any one of claims 1 to 4.

6. A computer (2) having a processor and a memory, The processor: Obtaining inspection result data for the objects to be inspected that are arranged on the production line from a plurality of inspection modules (50, 60, 70) that are arranged on the production line in a predetermined order; inputting the acquired inspection result data into a learning model associated with a first inspection module, and determining to which of a plurality of clusters the inspection object belongs based on an output from the learning model; acquiring setting values ​​of parameters used by inspection modules arranged in the production line other than the first inspection module, the setting values ​​being stored in the memory according to the cluster to which the object to be inspected belongs; computer.

7. 1. An orchestration program comprising: A computer (2) having a processor and a memory, a function of acquiring inspection result data for the objects to be inspected arranged on the production line from a plurality of inspection modules (50, 60, 70) arranged in a predetermined order on the production line; a function of inputting the acquired inspection result data into a learning model associated with a first inspection module, and determining to which of a plurality of clusters the inspection object belongs based on an output from the learning model; a function of acquiring setting values ​​of parameters used by inspection modules arranged on the production line other than the first inspection module, the setting values ​​being stored in the memory according to the cluster to which the inspection object belongs; An orchestration program that makes this possible.

8. An orchestration method by a computer (2) having a processor and a memory, comprising: the processor: Obtaining inspection result data for the objects to be inspected that are arranged on the production line from a plurality of inspection modules (50, 60, 70) that are arranged on the production line in a predetermined order; inputting the acquired inspection result data into a learning model associated with a first inspection module, and determining to which of a plurality of clusters the inspection object belongs based on an output from the learning model; acquiring setting values ​​of parameters used by inspection modules arranged in the production line other than the first inspection module, the setting values ​​being stored in the memory according to the cluster to which the object to be inspected belongs; Orchestration method.

Images