Analysis method, analysis device, program, and storage medium

Abstract

This invention provides an analysis method, analysis device, program, and storage medium that make it easier to understand the risks associated with components. [Solution] In the analysis method according to the embodiment, a computer is made to refer to a risk prediction model that has been trained using a plurality of first feature data and second feature data. The plurality of first feature data are each assigned to a plurality of risk items. The plurality of risk items include a first risk item related to a part, a second risk item related to a product, and a third risk item related to an implementation method. The second feature data indicates the risk for the plurality of first feature data. Third feature data assigned to each risk item for a new part is obtained. The third feature data is input to the risk prediction model, and fourth feature data indicating the overall risk for the third feature data is obtained from the risk prediction model. The analysis results using the fourth feature data are displayed.

Description

Technical Field

[0001] Embodiments of the present invention relate to an analysis method, an analysis apparatus, a program, and a storage medium.

Background Art

[0002] Many parts are used in products. In order to suppress the occurrence of defects in products, it is necessary to consider various risks when selecting parts.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The problem to be solved by the embodiments of the present invention is to provide an analysis method, an analysis apparatus, a program, and a storage medium that can more easily grasp the risks related to parts.

Means for Solving the Problems

[0005] In the analysis method according to the embodiment, the computer is made to refer to a risk prediction model that has been trained using a plurality of first feature data and second feature data. The plurality of first feature data are each assigned to a plurality of risk items. The plurality of risk items include a first risk item related to a component, a second risk item related to a product into which the component is incorporated, and a third risk item related to the mounting method of the component. The second feature data indicates the risk to the plurality of first feature data. Furthermore, the computer is made to acquire a plurality of third feature data that have been assigned to each of the plurality of risk items for a new component. Furthermore, the computer is made to input the plurality of third feature data into the risk prediction model and acquire a fourth feature data from the risk prediction model that indicates the overall risk to the plurality of third feature data. Furthermore, the computer is made to display the analysis results using the fourth feature data. [Brief explanation of the drawing]

[0006] [Figure 1] Figure 1 is a flowchart showing the analysis method according to the embodiment. [Figure 2] Figure 2 is a table showing an example of an analysis sheet. [Figure 3] Figure 3 is a table showing an example of training data. [Figure 4] Figure 4 is a table showing an example of the results of a risk prediction model. [Figure 5] Figure 5 is a graph showing an example of the analysis results. [Figure 6] Figure 6 is a chart showing another example of the analysis results. [Figure 7] Figure 7 is a schematic diagram representing the hardware configuration. [Modes for carrying out the invention]

[0007] The embodiments of the present invention will be described below with reference to the drawings. In this specification and in the drawings, elements similar to those already described are denoted by the same reference numerals, and detailed descriptions will be omitted as appropriate.

[0008] Generally, a single product uses many components. For example, if a product is an electronic device, various electronic components are used. Electronic components include passive components such as capacitors, resistors, transformers, and inductors, and active components such as diodes, transistors, and integrated circuits (ICs). Electronic devices may also include other components such as mechanical components (switches, connectors) and sensors.

[0009] To improve product quality, it is necessary to select components appropriately. For example, if a component has potential risks in its characteristics, is incompatible with other components, is not suited to the product's operating environment, or is not mounted correctly, component failures may occur during product operation. A high risk of component failure increases the risk of product failure. As a result, product quality or reliability may decrease. For example, if a component is responsible for power switching within a product, a failure in a single component could cause the entire product to stop working.

[0010] For example, if components are selected inappropriately, and a product is designed and developed using those components, and product defects become apparent after development, the design and development will need to be redone. If defects become apparent after the product has been sold, product recalls and defect resolution will be necessary. For this reason, it is desirable to select components more appropriately during the product design phase.

[0011] When selecting components, various risks must be considered. However, selecting components while considering risks requires sufficient experience and knowledge. Inexperienced engineers find it difficult to select appropriate components. Because only a limited number of people are capable of selecting components, the selection process sometimes takes a long time.

[0012] Embodiments of the present invention are used to investigate the risks of a new component when its adoption is being considered.

[0013] Figure 1 is a flowchart showing the analysis method according to the embodiment. At least a portion of the analysis method according to the embodiment is performed by a computer. A general-purpose computer can be used for the analysis.

[0014] In the analysis method shown in Figure 1, the computer first obtains an analysis sheet for the new part (step S1). The analysis sheet is prepared in advance by a person (e.g., a designer). The analysis sheet may be entered by the computer user, or the computer may automatically obtain it from a storage area such as a server.

[0015] Figure 2 is a table showing an example of an analysis sheet. The analysis sheet 100 shown in Figure 2 includes Name 101, Part Number 102, Specifications 103, and Risk Item Group 110. Name 101 indicates the name of each part. Part Number 102 is a string used to identify each part. Specifications 103 indicates the shape, structure, dimensions, composition, precision, performance, operating conditions, etc., of each part.

[0016] Risk item group 110 includes the first risk item group 111, the second risk item group 112, and the third risk item group 113. The first risk item group 111 includes one or more risk items indicating risks related to the component. The second risk item group 112 includes one or more risk items indicating risks related to the product into which the component is incorporated. The third risk item group 113 includes one or more risk items indicating risks related to the method of mounting the component. Each risk item indicates a potential risk factor and is defined based on historical knowledge of the component, product, and mounting method.

[0017] In the illustrated example, the first risk item group 111 includes four first risk items 111a to 111d. The second risk item group 112 includes four second risk items 112a to 112d. The third risk item group 113 includes four third risk items 113a to 113d.

[0018] For each risk item, for each product, feature amount data (third feature amount data) indicating the presence or absence of risk is described. The feature amount data is pre-input by a person. The person pre-assigns the feature amount data for each risk item within the known range for the new component.

[0019] In the example shown in FIG. 2, the new component is a resistor. The specifications of each of the four resistors R1 to R4 and the risk items related to the resistors are described. The feature amount data is either "1", "0", or "-1". "1" indicates that there is no risk. "0" indicates that it is unknown. "-1" indicates that there is a risk.

[0020] In the illustrated example, "1" is input when the content described in the risk item is "YES", and "-1" is input when it is "NO". For example, regarding the first risk item 111c, it shows that "Jumper resistor" contains lead in the electrode plating, while the electrode plating of other resistors does not contain lead. Regarding the third risk item 113b, it shows that nitrogen reflow is not used in the mounting of "surge-resistant resistor", while nitrogen reflow is used in the mounting of other resistors.

[0021] Once the analysis sheet is obtained, the computer references the risk prediction model M (step S2). The risk prediction model is pre-trained to output feature data that represents the overall risk of a part, in response to input of multiple feature data. Past product and part information is used as training data. For example, the risk prediction model is machine-learned using logistic regression or a decision tree. The risk prediction model may be stored on the computer, or the computer may reference a risk prediction model stored on a server via a network.

[0022] Figure 3 is a table showing an example of training data. As shown in Figure 3, the learning data 200 includes name 201, part number 202, specifications 203, input data 210, and judgment flag 220. Name 201 indicates the name of each part. Part number 202 is a string used to identify each part. Specifications 203 indicates the shape, structure, dimensions, composition, precision, performance, operating conditions, etc., of each part.

[0023] The input data 210 includes multiple feature data (first feature data). The multiple feature data in the input data 210 are assigned to the same multiple risk items as in the analysis sheet 100. The judgment flag 220 includes feature data (second feature data) that indicates the overall risk of each component. The training data 200 is created by analyzing past components.

[0024] For each component, a group of input data 210 and a judgment flag 220 is created. In the training data 200, multiple resistors r1~r n For each of these, a set of input data 210 and judgment flags 220 are provided. The risk prediction model M is machine-learned to associate multiple feature data of the input data 210 with the feature data of the judgment flags 220 so that it outputs feature data.

[0025] The computer inputs the analysis sheet into the risk prediction model (Step S3). The computer obtains the judgment result from the risk prediction model (Step S4). The judgment result includes feature data (fourth feature data), which indicates the overall risk of the component.

[0026] Figure 4 is a table showing an example of the results of a risk prediction model. Table 300, shown in Figure 4, includes the data from the analysis sheet 100 shown in Figure 2, as well as the judgment results 310 from the risk prediction model. In the illustrated example, the judgment result 310 includes one of the following feature data as the overall risk of the part: "1" (no risk), "0" (unknown), or "-1" (risk present). The judgment result 310 may further include the probability (certainty) of each judgment result: "1", "0", and "-1".

[0027] Once the judgment result is obtained from the risk prediction model, the computer performs an analysis using the judgment result (step S5). The computer displays the analysis result on a display device (step S6).

[0028] For example, the computer references multiple first feature data contained in the input data 210 of the training data 200. The computer dimensionally compresses the multiple first feature data assigned to one part to generate two-dimensional or three-dimensional compressed first data. The computer dimensionally compresses each of the multiple first feature data assigned to each part. This generates multiple compressed first data.

[0029] Furthermore, the computer acquires multiple third-party feature data included in the risk item group 110 of the analysis sheet 100. The computer then compresses the dimensionality of the multiple third-party feature data assigned to a single new component to generate two-dimensional or three-dimensional second-party compressed data. Methods such as t-Distributed Stochastic Neighbor Embedding (t-SNE) or Principal Component Analysis (PCA) can be used for dimensionality reduction.

[0030] Figure 5 is a graph showing an example of the analysis results. Figure 5 shows the results of compressing and plotting multiple first-order feature data and multiple third-order feature data in two dimensions. Figure 5 is an example of the first analysis result. In the analysis result 400 shown in Figure 5, multiple first-order compressed data 401 related to past parts and multiple second-order compressed data 402 related to new parts are plotted.

[0031] As shown in Figure 5, the computer may change the display of the first compressed data 401 according to the second feature data (-1, 0, or 1) of the decision flag 220. In the illustrated example, the color of the first compressed data 401 changes according to the second feature data of the decision flag 220. Specifically, the colors of the first compressed data 401a, the first compressed data 401b, and the first compressed data 401c are different from each other. The first compressed data 401a is data obtained by dimensionality reduction of multiple first feature data for which the corresponding second feature data is "-1". The first compressed data 401b is data obtained by dimensionality reduction of multiple first feature data for which the corresponding second feature data is "0". The first compressed data 401c is data obtained by dimensionality reduction of multiple first feature data for which the corresponding second feature data is "1". Instead of, or in addition to, the shape or size of the first compressed data 401 may change according to the second feature data of the decision flag 220.

[0032] The computer may change the display of the second compressed data 402 according to the fourth feature data (-1, 0, or 1) of the determination result. In the illustrated example, the color of the second compressed data 402 changes according to the fourth feature data of the determination result. Specifically, the colors of the second compressed data 402a, the second compressed data 402b, and the second compressed data 402c are different from each other. The second compressed data 402a is data obtained by dimensionality reduction of multiple third feature data whose corresponding fourth feature data is "-1". The second compressed data 402b is data obtained by dimensionality reduction of multiple third feature data whose corresponding fourth feature data is "0". The second compressed data 402c is data obtained by dimensionality reduction of multiple third feature data whose corresponding fourth feature data is "1". Instead of, or in addition to, the shape or size of the second compressed data 402 may change according to the fourth feature data of the determination result.

[0033] Furthermore, the computer may display the first compressed data 401 and the second compressed data 402 differently. In the illustrated example, the shapes of the first compressed data 401 and the second compressed data 402 are different. By differentiating one or more elements selected from color, size, and shape depending on the data, the user can better understand the analysis results.

[0034] As shown in the analysis results in Figure 5, users can easily understand the relationship between the characteristics of past parts and the characteristics of new parts. For example, users can easily understand the relationship between the judgment result of new parts and parts that have had risks in the past and parts that have not had risks in the past. In the analysis results, if the plot of new parts is close to the group of plots of parts that had risks, users can understand that there is a higher probability that risks will occur with the new parts.

[0035] Figure 6 is a chart showing another example of the analysis results. Figure 6 shows decision tree 500 based on the feature data of analysis sheet 100. Figure 6 is an example of the second analysis result. A decision tree is a machine learning method that uses a tree structure to perform classification or regression. For machine learning analysis, algorithms in the Python library, an open-source programming language for analysis and modeling, can be used. For example, the data from analysis sheet 100 is processed in Python. At this time, the results are weighted and selected using a machine learning-prepared risk prediction model M, and then visualized using the export_graphviz function to generate decision tree 500 shown in Figure 6.

[0036] In the illustrated example, decision tree 500 contains multiple nodes 501-511. Each of nodes 501-511 corresponds to one of the multiple risk factors listed in analysis sheet 100.

[0037] In the illustrated example, each node includes the splitting criterion, Gini impurity (gini), sample size, class distribution (value), and dominant class. The splitting criterion indicates how the node divides the data. Gini impurity indicates the degree of class mixing within the node. Sample size indicates the number of data points handled by that node. Class distribution indicates the class distribution of the data within the node. The dominant class indicates the most prevalent class within the node. From each node, data branches according to the feature data for the risk factors in analysis sheet 100.

[0038] By outputting the analysis results shown in Figure 6, users can more easily understand the basis for the judgment results from the risk prediction model. For example, if the risk prediction model outputs a judgment result of "-1" (risk present), users can more easily understand why it was judged as having a risk, and which risk items need to be improved to potentially result in a judgment of no risk.

[0039] The advantages of the embodiment will be explained. As described above, in order to suppress the occurrence of defects in products, it is necessary to consider various risks when selecting parts. However, selecting parts while considering risks is difficult for inexperienced engineers. In this embodiment of the present invention, a risk prediction model is used to address this problem. The risk prediction model is learned using multiple first feature data assigned to multiple risk items, and second feature data indicating the risk for those multiple first feature data. For example, the risk prediction model is constructed by machine learning using information on past products and parts. By inputting multiple third feature data assigned to multiple risk items related to a new part into this risk prediction model, a fourth feature data indicating the overall risk of that new part can be obtained. The computer also analyzes the fourth feature data and displays various analysis results. By referring to the analysis results, the user can easily grasp the risks related to the new part.

[0040] In particular, electronic equipment and control boards for infrastructure applications incorporate a wide variety of electronic components. Furthermore, infrastructure products often require long-term reliability. If critical components in electronic equipment or control boards malfunction, it can lead to product malfunctions or even complete shutdown. This can disrupt daily life and have a significant impact. Therefore, for infrastructure products in particular, it is crucial that no malfunctions occur within the product's lifecycle. However, in recent years, substandard and counterfeit parts have entered the market, and the incorrect selection of these parts can trigger major malfunctions in infrastructure products during operation. For this reason, there is a need for technology that can predict and visualize potential future risks using component information, manufacturing information, and product operation information from the upstream design phase of the product being designed.

[0041] According to this embodiment, it becomes possible to easily identify the risks associated with new components. For example, when selecting electronic components to be used in infrastructure products, the risks of new electronic components can be identified and utilized in the design. As a result, the occurrence of defects in infrastructure products can be suppressed.

[0042] Figure 7 is a schematic diagram representing the hardware configuration. The analysis method according to this embodiment is performed by a computer. For example, as shown in Figure 7, the computer 90 includes a processing circuit 91, a ROM 92, a RAM 93, a storage device 94, an input interface 95, an output interface 96, and a communication interface 97.

[0043] ROM92 stores programs that control the operation of computer 90. ROM92 contains the programs necessary for computer 90 to perform each of the processes described above. RAM93 functions as a memory area where the programs stored in ROM92 are loaded.

[0044] The processing circuit 91 includes an arithmetic processing unit (CPU, GPU, etc.). The processing circuit 91 uses RAM 93 as work memory and executes a program stored in at least one of ROM 92 or storage device 94. During program execution, the processing circuit 91 controls each component via the system bus 98 and performs various processes.

[0045] The memory device 94 stores data necessary for program execution and data obtained through program execution.

[0046] The input interface (I / F) 95 can connect the computer 90 and the input device 95a. The input I / F 95 is, for example, a serial bus interface such as USB. The processing circuit 91 can read various data from the input device 95a via the input I / F 95.

[0047] The output interface (I / F) 96 can connect the computer 90 and the output device 96a. The output I / F 96 is a video output interface such as a Digital Visual Interface (DVI) or a High-Definition Multimedia Interface (HDMI®). The processing circuit 91 can transmit data to the output device 96a via the output I / F 96 and display an image on the output device 96a.

[0048] The communication interface (I / F) 97 can connect the computer 90 to a server 97a located outside the computer 90. The communication I / F 97 is, for example, a network module such as a LAN module. The processing circuit 91 can read various data from the server 97a via the communication I / F 97.

[0049] The storage device 94 includes one or more selected from Hard Disk Drives (HDDs) and Solid State Drives (SSDs). The input device 95a includes one or more selected from a mouse, keyboard, microphone (voice input), and touchpad. The output device 96a includes one or more selected from a monitor, projector, printer, and speaker. Devices that have the functions of both input device 95a and output device 96a, such as a touch panel, may also be used.

[0050] For example, by having a computer execute the analysis method according to the embodiment, an analysis device capable of analyzing risks related to parts can be obtained. The processing required for the analysis method according to the embodiment may be implemented by a single computer 90, or by the cooperation of multiple computers 90. The multiple computers 90 may be connected to each other via a network.

[0051] The processing of the various data described above may be recorded as a program that can be executed by a computer on a magnetic disk (flexible disk and hard disk, etc.), an optical disk (CD-ROM, CD-R, CD-RW, DVD-ROM, DVD±R, DVD±RW, etc.), a semiconductor memory, or another non-transitory computer-readable storage medium.

[0052] For example, data on a recording medium is read by a computer (or embedded system). The recording format (storage format) on the recording medium is arbitrary. For example, a computer reads a program from the recording medium and causes the CPU to execute instructions based on this program. The acquisition (or reading) of the program by the computer may be done via a network.

[0053] Embodiments of the present invention include the following features. (Feature 1) On the computer, A risk prediction model trained using multiple first feature data, each assigned to a plurality of risk items including a first risk item related to a component, a second risk item related to the product in which the component is incorporated, and a third risk item related to the method of mounting the component, and second feature data indicating the risk for the plurality of first feature data, is referenced. For the new component, obtain multiple third feature data assigned to each of the aforementioned multiple risk items. The plurality of third feature data are input into the risk prediction model, and a fourth feature data representing the overall risk for the plurality of third feature data is obtained from the risk prediction model. Display the analysis results using the aforementioned four feature data. Analysis method. (Feature 2) In displaying the analysis results, the computer, By compressing the dimensionality of the aforementioned multiple first feature data, the first compressed data is generated. By compressing the dimensions of the aforementioned multiple third feature data, a second compressed data is generated. Display the first analysis result, which includes the first compressed data and the second compressed data. The analysis method described in Feature 1. (Feature 3) The risk prediction model is trained using multiple sets of the multiple first feature data and the multiple second feature data. In the aforementioned first analysis results, multiple previously compressed data sets are displayed. The analysis method according to Feature 2, wherein the display of each of the plurality of first compressed data differs according to the corresponding second feature data. (Feature 4) The analysis method according to any one of features 1 to 3, wherein, in displaying the analysis results, the computer displays a second analysis result that includes a decision tree showing at least a portion of the plurality of risk items and at least a portion of the plurality of third feature data. (Feature 5) Equipped with a processing circuit, An analytical device that performs the analytical method described in one of the features 1 to 4. (Feature 6) A program that causes a computer to perform an analysis method described in one of the features 1 to 4. (Feature 7) A storage medium containing the program described in Feature 6.

[0054] According to the embodiments described above, an analysis method, an analysis apparatus, a program, and a storage medium are provided that make it easier to understand the risks related to the components.

[0055] In this specification, "or" indicates that "at least one" of the items listed in the text may be adopted.

[0056] Although several embodiments of the present invention have been illustrated above, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. Furthermore, the embodiments described above can be implemented in combination with each other. [Explanation of Symbols]

[0057] 100: Analysis sheet, 110: Risk item group, 111: First risk item group, 112: Second risk item group, 113: Third risk item group, 200: Training data, 210: Input data, 220: Judgment flag, 300: Table, 310: Judgment result, 400: Analysis result, 401, 401a~401c: First compressed data, 402, 402a~402c: Second compressed data, 500: Decision tree

Claims

1. On the computer, A risk prediction model trained using a plurality of first feature data, each assigned to a plurality of risk items including a first risk item related to a component, a second risk item related to the product in which the component is incorporated, and a third risk item related to the method of mounting the component, and a second feature data indicating the risk for the plurality of first feature data, is referenced. Regarding the new component, obtain multiple third-feature data assigned to each of the aforementioned multiple risk items. The plurality of third feature data are input into the risk prediction model, and a fourth feature data representing the overall risk for the plurality of third feature data is obtained from the risk prediction model. Display the analysis results using the aforementioned four feature data. Analysis method.

2. In displaying the analysis results, the computer, By compressing the dimensions of the aforementioned plurality of first feature data, first compressed data is generated. By compressing the dimensions of the aforementioned multiple third feature data, a second compressed data is generated. Display the first analysis result, which includes the first compressed data and the second compressed data. The analysis method according to claim 1.

3. The risk prediction model is trained using multiple sets of the multiple first feature data and the multiple second feature data. In the first analysis result, multiple first compressed data are displayed. The analysis method according to claim 2, wherein the display of each of the plurality of first compressed data differs according to the corresponding second feature data.

4. The analysis method according to claim 1, wherein, in displaying the analysis results, the computer displays a second analysis result that includes a decision tree showing at least a portion of the plurality of risk items and at least a portion of the plurality of third feature data.

5. Equipped with a processing circuit, An analysis apparatus for performing the analysis method described in any one of claims 1 to 4.

6. A program that causes a computer to perform the analysis method described in any one of claims 1 to 4.

7. A storage medium storing the program described in claim 6.

Images