Inference device, information processing device, inference method, program, and recording medium

Abstract

The present invention is an inference device for inferring a phenomenon, comprising: a question acquisition unit, which acquires a question related to the phenomenon; a question determination unit, which determines whether the question is a qualitative question or a quantitative question; a sensor determination unit, which determines whether sensor data can be obtained in the case of a quantitative question; a decision unit, which determines the sensor data as data for inference if the sensor data can be obtained, and determines user-based input data as data for inference if the sensor data cannot be obtained; and an inference unit, which uses the data determined by the decision unit to perform inference corresponding to the phenomenon.

Description

Technical Field

[0001] The present invention relates to an inference device, an information processing device, an inference method, a program, and a recording medium for performing inference based on input information. Background Art

[0002] Conventionally, there are known expert systems that infer factors causing defects in manufacturing processes or infer countermeasures corresponding to the defects. In the expert system, predetermined questions are asked to the user, and the factors and countermeasures are inferred based on the user's answers to the questions.

[0003] Due to recent technological advances, sensor input data has become increasingly large, making it possible to use more sensor data even in factor inference. Patent Document 1 discloses a device that performs appropriate factor inference by determining whether to use user input data or sensor data as an answer.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2007-279840 Summary of the Invention

[0007] Problems to be solved by the invention

[0008] However, the technology disclosed in Patent Document 1 requires knowing whether a device to be monitored has a sensor, and presetting which of the input data input from the user and the sensor data is to be prioritized according to the device to be monitored, which is time-consuming.

[0009] The present invention has been made in view of the above-mentioned problems, and an object of the present invention is to provide an inference device that can appropriately infer a phenomenon at low cost.

[0010] Means for solving problems

[0011] Therefore, the present invention is an inference device for inferring a phenomenon, characterized in that it comprises: a problem acquisition unit, which acquires a problem related to the phenomenon; a problem determination unit, which determines whether the problem is a qualitative problem or a quantitative problem; a sensor determination unit, which determines whether sensor data can be obtained in the case of the quantitative problem; a decision unit, which determines the sensor data as data for inference when the sensor data can be obtained, and determines user-based input data as data for inference when the sensor data cannot be obtained; and an inference unit, which uses the data determined by the decision unit to perform inference corresponding to the phenomenon.

[0012] Effects of the Invention

[0013] According to the present invention, it is possible to provide an inference device that can appropriately infer a phenomenon without incurring any cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is the overall structural diagram of the inference system.

[0015] Figure 2 It is a hardware structure diagram of the inference device.

[0016] Figure 3 It is a functional structure diagram of the inference device.

[0017] Figure 4 This is a diagram showing an example of the data structure of the question DB.

[0018] Figure 5 This is a diagram showing an example of the data structure of a link DB.

[0019] Figure 6 This is a diagram of the data structure of the candidate DB.

[0020] Figure 7 It is a conceptual diagram of the knowledge database.

[0021] Figure 8 This is the data structure diagram of the word DB.

[0022] Figure 9 This is a flowchart showing the inference processing performed by the inference device.

[0023] Figure 10 This is a flowchart showing the inference processing performed by the inference device.

[0024] Figure 11 A diagram showing a display example during inference processing.

[0025] Figure 12 This is a diagram showing a display example during inference processing.

[0026] Figure 13A This is a diagram showing a display example during inference processing.

[0027] Figure 13B This is a diagram showing a display example during inference processing.

[0028] Figure 13C This is a diagram showing a display example during inference processing. DETAILED DESCRIPTION

[0029] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0030] Figure 1This is a diagram of the overall structure of an inference system. An inference system infers factors, such as the underlying causes, from phenomena. This embodiment uses the example of a phenomenon, such as a failure or anomaly, occurring in a monitored device, and the inference system inferring countermeasures to that phenomenon. The phenomenon is not limited to failures in the monitored device. Furthermore, the inference target is not limited to countermeasures; it can also be factors underlying the phenomenon.

[0031] The inference system includes an inference device 100, a device 110 to be monitored, a sensor group 120, and smart glasses 130. The sensor group 120 includes multiple sensors that detect various information about the device 110 to be monitored. The inference device 100 is configured as, for example, a server or cloud-based information processing device, and is communicatively connected to the sensor group 120 and smart glasses 130 via, for example, a network. The inference device 100 obtains various sensor data from the sensor group 120. The smart glasses 130 are wearable devices that display images superimposed on real space. The smart glasses 130 receive inference results and other information from the inference device 100 and display them. The smart glasses 130 also transmit user input to the inference device 100. The inference device 100 uses the sensor data and input data entered by the user into the smart glasses 130 to infer countermeasures for the device 110. The device 110 can be any device capable of inferring countermeasures using both sensor data and input data, and its type is not particularly limited. Examples of device 110 include DC coolers, etc. Inference apparatus 100 is not limited to being connected to a single device 110 and its sensor group 120, but may be communicatively connected to multiple devices and sensor groups to infer phenomena related thereto.

[0032] Figure 2 This is a hardware configuration diagram of the inference device 100. The inference device 100 includes a CPU 201, a ROM 202, a RAM 203, an HDD 204, a display unit 205, an input unit 206, and a communication unit 207. The CPU 201 reads the control program stored in the ROM 202 and executes various processes. The RAM 203 is used as a temporary storage area, such as the main memory and work area of the CPU 201. The HDD 204 stores various data, various programs, etc. The display unit 205 displays various information. The input unit 206 includes a keyboard and a mouse and accepts various operations performed by the user. The communication unit 207 performs communication processing with external devices such as sensors via a network.

[0033] The functions and processing of the inference device 100 described below are implemented by CPU 201 reading and executing a program stored in ROM 202 or HDD 204. Alternatively, CPU 201 may read a program stored in a recording medium such as an SD card instead of ROM 202.

[0034] Figure 3 3 is a functional configuration diagram of the inference device 100. The inference device 100 includes a question DB 301, a link DB 302, a candidate DB 303, a word DB 304, a sensor data DB 305, an input management unit 311, and an inference unit 312.

[0035] Figure 4 This is a diagram showing an example of the data structure of question DB 301. Question DB 301 stores question data, costs, and links in association with each other. Here, question data is question data required to infer countermeasures. Cost is an indicator value that represents the size of the burden associated with answering question data. For example, a question that cannot be answered without temporarily stopping a process such as continuous casting will be accompanied by a loss in answering. A high cost is set for such question data. Cost data is used as a judgment factor when selecting question data, and question data with high cost data is difficult to select. This can prevent an increase in cost. Links are information that associate candidate data described later with question data.

[0036] Figure 5 This diagram shows an example of the data structure of link DB 302. Link DB 302 stores links, candidate data, and influence in an associated manner. Here, candidate data refers to data that is a candidate for the inference result of the inference unit 312. Influence is assigned as positive or negative data depending on the relationship between the question data and the candidate data. For example, in the case of a question used to select one of two candidate data, one candidate data is assigned data that is positive relative to the question, while the other candidate data is assigned data that is negative relative to the question.

[0037] Figure 6 This is a diagram of the data structure of candidate DB 303. Candidate DB 303 stores candidate data in association with confidence. Here, confidence is a value indicating the certainty of the inference result for the candidate data. Confidence is initially set to 50% for all cases and is updated as the inference process progresses.

[0038] Figure 7 This is a conceptual diagram of a knowledge database implemented by question DB 301, link DB 302, and candidate DB 303. In this way, question data (Q1, Q2, ...) are associated with candidate data (N1 to N21) through links (L1, L2, ...). Figure 7As shown, the candidate data are hierarchically arranged in a tree. In addition, each question data is associated with the candidate data regardless of its hierarchy. In other words, multiple candidate data of different hierarchies can be associated with one question data.

[0039] Figure 8 This is a data structure diagram of word DB304. Word DB304 associates the words included in the question with the type of question and stores them. Here, as the type of question, there are two types: qualitative questions and quantitative questions. Here, the so-called quantitative questions are questions such as whether the temperature of the device is within the range of 10°C to 20°C, and the answers to which can be obtained as sensor data. On the other hand, the so-called qualitative questions are questions such as whether the device is contaminated, and the answers to which cannot be obtained as sensor data. In addition, the type of question is pre-set by word. And, information indicating the type of sensor data is associated with the quantitative question. The so-called type of sensor data is the type of value such as temperature, humidity, etc. In addition, word DB304 is an example of a correspondence table.

[0040] Return to Figure 3 The inference unit 312 selects question data by referring to the question DB 301, the link DB 302, and the candidate DB 303, and infers a countermeasure based on the answer data obtained for the question data. Based on the question data selected by the inference unit 312, the input management unit 311 determines whether the data used for inference should be input data entered by the user or sensor data input from the sensor group 120. The input management unit 311 then transmits the determined data to the inference unit 312. When determining the data used for inference based on the question data, the input management unit 311 refers to the word DB 304. The sensor data DB 305 stores the sensor data input from the sensor group 120.

[0041] Figure 9 and Figure 10 1 is a flowchart showing the inference processing performed by the inference device 100 . Figure 11 、 Figure 12 as well as Figures 13A to 13C 1 is a diagram showing an example of a display of the smart glasses 130 during inference processing. Figure 11 As shown, when an abnormality occurs, information indicating the source of the abnormality is displayed on the smart glasses 130 as in display example 1101. In display example 1101, it is displayed as a cooler. And, "1" is displayed in the cooler. When the user selects "1" by saying "1", information indicating the operation for the cooler is displayed as shown in display example 1102. Here, when the user selects "1", fault diagnosis (inference processing) starts. In addition, in Figure 11 、 Figure 12 as well as Figures 13A to 13CIn the example shown, for the sake of convenience, the real space that the user can visually confirm is omitted and only the image displayed in an overlapping manner is shown. However, in reality, the user wearing the smart glasses 130 can observe the real space in a state of overlapping with the real space. Figure 11 、 Figure 12 as well as Figures 13A to 13C The image displayed.

[0042] In the inference process, first, in S901, the inference unit 312 selects one arbitrary question data from the question DB 301. Next, in S902, the inference unit 312 determines whether an answer has been received for the selected question data. If an answer has been received (S902: Yes), the inference unit 312 proceeds to S904. If no answer has been received (S902: No), the inference unit 312 proceeds to S903.

[0043] In S903, the inference unit 312 calculates the appropriate value of the question data. Specifically, the inference unit 312 calculates the appropriate value by (Formula 1). The inference unit 312 finds the effect by (Formula 2). Here, the cost is the cost for the question data being selected. The effect is the effect of the question data being selected. The influence and the confidence are both the influence and confidence corresponding to the candidate data associated with the question data being selected via a link. In the case of corresponding to multiple candidate data, the influence and confidence corresponding to the multiple candidate data are used in (Formula 2). In addition, the absolute value is used for the influence.

[0044] Appropriate value = cost × effect (Formula 1)

[0045] Effect = Sum of (Each Impact × Each Confidence) (Formula 2)

[0046] In S904, the inference unit 312 determines whether the calculation of appropriate values for all question data has been completed. If the processing for all question data has been completed (S904: Yes), the inference unit 312 advances the process to S905. If unprocessed question data exists (S904: No), the inference unit 312 advances the process to S901. In this case, in S901, unprocessed question data is again selected for subsequent processing.

[0047] In S905 , the inference unit 312 selects the best question data based on the appropriate value. Specifically, the inference unit 312 selects the question data with the largest appropriate value. The inference unit 312 then transmits the selected question data to the input management unit 311 .

[0048] Next, in S906, the input management unit 311 obtains question data from the inference unit 312 and determines the type of the obtained question data. Specifically, the input management unit 311 extracts the words included in the question data. Then, the input management unit 311 refers to the word DB 304 and determines whether the type associated with the words included in the question data is qualitative or quantitative. In addition, when multiple word data are extracted from the question data, the input management unit 311 determines the type of question data from the multiple words according to predetermined conditions. In addition, the input management unit 311 only needs to determine the type based on the question data, and the specific process for determining the type is not limited to the embodiment. The process of S906 is an example of the question acquisition process and the question determination process. In the case of a quantitative question (S906 is quantitative), the input management unit 311 advances the process to S907. In the case of a qualitative question (S906 is qualitative), the input management unit 311 advances the process to S910.

[0049] In S907, the input management unit 311 further refers to the word DB 304 to determine the type of sensor data to be obtained. This process is an example of sensor determination processing. Then, the input management unit 311 determines whether the sensor data of the determined type can be obtained. In addition, the input management unit 311 determines the type of sensor data that can be obtained based on the sensor data input from the sensor group 120. In addition, the process of S907 is an example of sensor determination processing. When the sensor data of the determined type can be obtained (S907 Yes), the input management unit 311 advances the process to S908. When the sensor data of the determined type cannot be obtained (S907 No), the input management unit 311 advances the process to S910.

[0050] In S908, the input management unit 311 acquires the specified type of sensor data. This process is an example of sensor data acquisition. Furthermore, the input management unit 311 determines whether the sensor data is normal. For example, in temperature data, if the expected detection range is 10-20°C and a value of -10°C is detected, it is considered that the value is not correct due to a sensor anomaly or other reason. The process in S908 eliminates such unexpected values.

[0051] Specifically, the input management unit 311 determines whether the sensor data obtained from the sensor group 120 is normal data according to pre-set conditions for each type of sensor data. For example, if an allowable range of 10 to 30°C is defined for temperature data, the input management unit 311 determines that the obtained sensor data is normal data if it is within the allowable range, and determines that it is not normal data if it is outside the allowable range. Alternatively, as another example, the input management unit 311 may determine whether the data is normal based on the time series changes in the temperature data detected at the time of processing. For example, the input management unit 311 predicts the value of the next sensor data based on the time series changes. Alternatively, the obtained sensor data may be determined to be normal data if it is within a predetermined range from the predicted value, and determined to be not normal data if it is not within the predetermined range. Furthermore, the process in S908 is an example of data determination processing.

[0052] If the data is normal (S908: Yes), the input management unit 311 advances the process to S909. If the data is not normal (S908: No), the input management unit 311 advances the process to S910.

[0053] In S909, the inference unit 312 generates answer data based on the sensor data acquired in S908. In this embodiment, the question data can be answered with any of the following: yes, no, or unknown. The input management unit 311 generates any of the following: yes, no, or unknown as the answer data from the sensor data. After processing in S909, the input management unit 311 proceeds to S912.

[0054] On the other hand, in S910, the input management unit 311 controls the output of the question data to the smart glasses 130 via the communication unit 207. Figure 12 In the example display 1201 shown, the question "Is the steam flow rate 1L or more or 3L or less per minute?" is displayed. The user then enters an answer to the question. At S911, the input management unit 311 receives the user's input data (answer data) via the communication unit 207. Here, the received answer data is also one of yes, no, and unknown, as described above. After processing S911, the input management unit 311 proceeds to S912.

[0055] In S912, the input management unit 311 updates the confidence level of each of the multiple candidate data associated with the selected question data based on the answer data obtained in S909 or S911. Specifically, if the answer data is yes, the input management unit 311 increases the confidence level of all candidate data associated with the question data by a predetermined amount. On the other hand, if the answer data is no, the input management unit 311 subtracts the confidence level of all candidate data associated with the question data by a predetermined amount. In the case of unknown answers, there is no change in the confidence level. The processing in S912 is an example of inference processing.

[0056] At S913, the input management unit 311 determines whether the confidence level of the predetermined candidate data has decreased. The processes from S901 to S915 are iterative, repeatedly updating the confidence level in accordance with the response data. This results in the confidence level of candidate data closer to the solution gradually increasing. Therefore, if the confidence level decreases after increasing to a certain level, it is considered that the response data may contain an error. The process of S913 determines the possibility of error in the response data.

[0057] The input management unit 311 selects candidate data that meets pre-set conditions, such as candidate data with the maximum confidence level or candidate data with a confidence level above a threshold, as processing targets. Furthermore, the input management unit 311 determines whether the confidence level calculated in S912 for the candidate data being processed has decreased compared to the confidence level before calculation. If the confidence level has decreased (Yes in S913), the input management unit 311 proceeds to S914. If the confidence level has not decreased (No in S913), the input management unit 311 proceeds to S915.

[0058] In S914, the input management unit 311 associates the information indicating that the answer data may be wrong with the question data being selected and records it. In addition, the input management unit 311 controls the display unit 205 to display information indicating that the answer data may be wrong. In addition, when the update is performed in S912 in a manner that reduces the confidence level, that is, when the update is performed in a manner that reduces the confidence level based on the sensor data, the sensor data may be abnormal. The input management unit 311 may also record and display the information in a manner that can identify the possibility of an abnormality in the sensor data. After the processing of S914, the input management unit 311 advances the processing to S915. In addition, the processing of displaying the information indicating the possibility of an error may be omitted.

[0059] In S915, the input management unit 311 determines whether the estimation is complete. If the maximum confidence value is greater than a preset threshold, the input management unit 311 determines that the estimation is complete. If the estimation is complete (Yes in S915), the input management unit 311 proceeds to S916. If the estimation is not complete (No in S915), the input management unit 311 proceeds to S901.

[0060] In S916, the input management unit 311 sends the estimation result to the smart glasses 130 via the communication unit 207. This process is an example of an output process of outputting the estimation result. The smart glasses 130 displays the estimation result when receiving the estimation result. As the estimation result, the input management unit 311 sends the candidate data with a confidence level above the threshold value together with the confidence level to the smart glasses 130. When there are multiple candidate data with a confidence level above the threshold value, the input management unit 311 sends a predetermined number of candidate data and their corresponding confidence levels to the smart glasses 130 in order from the highest level. Correspondingly, as Figure 12 As in the display example 1202 of FIG. 1 , candidate data with high confidence levels are displayed together with the confidence levels on the smart glasses 130. In the display example 1202, candidate data up to the third rank are displayed.

[0061] Then, in Figure 10 In S1001 shown, the inference unit 312 determines whether a display instruction has been received from the smart glasses 130. For example, when the user says Figure 12 In the case of "0" corresponding to "Confirm diagnosis process" shown in the display example 1202, a process display instruction is sent from the smart glasses 130 to the inference device 100. Here, the so-called process display instruction is information indicating the display of the diagnosis process information. The so-called diagnosis process information is the questions and their answers used in the process of inferring the diagnosis results, as well as information represented in a time series. When the inference unit 312 accepts the process display instruction (S1001 is), the processing proceeds to S1002. When the inference unit 312 does not accept the process display instruction (S1001 is not), the inference processing ends.

[0062] In S1002 , the inference unit 312 outputs diagnosis process information. Figure 13A Display example 1301 showing information on the diagnosis process. In display example 1301, questions 1 to 6 and their answers are displayed. Among them, in questions 4 to 6, "(automatically answered)" is displayed after the question text. This means that the inference device 100 automatically obtains sensor data as an answer without the user's confirmation. In this way, the user can confirm the questions and their answers using sensor data. In addition, in question 6, it is displayed as "*There is a possibility of an abnormal value!". In reference Figure 9In the processing of S914 described above, the answer data in which the possibility of error is recorded is displayed in a user-recognizable manner, thereby enabling the user to confirm whether the answer data is correct.

[0063] Here, if the user confirms the answer to question 6 and determines that there is an error in the input of sensor data, it can be changed through user operation. For example, suppose the user determines that the answer to question 6, humidity 20%, is incorrect. In this case, "6" is spoken. Correspondingly, Figure 13B As shown in the display example 1302, the smart glasses 130 display a window 1303 for inputting the changed value. Here, the user inputs the changed value. For example, the input is 60%. Correspondingly, Figure 13C As shown in display example 1304 of , the smart glasses 130 update the answer to question 6. Afterwards, when the user says "re-diagnose", the smart glasses 130 send a sensor data change instruction to the inference device 100, indicating that the answer to question 6 will be changed to 60%.

[0064] Accordingly, in S1003, the inference unit 312 determines whether a sensor data change instruction has been accepted. If the sensor data change instruction has been accepted (S1003: Yes), the inference unit 312 proceeds to S1004. If the sensor data change instruction has not been accepted (S1003: No), the inference unit 312 terminates the inference process. In S1004, the inference unit 312 changes the sensor data in accordance with the sensor data change instruction. For example, if the sensor data change instruction has been accepted, the sensor data serving as the answer to question 6 is changed from 20% to 60%. The sensor data change instruction indicates that the answer to question 6 is changed to 60%.

[0065] Next, in S1005, the inference unit 312 updates the confidence level for the candidate data associated with the corresponding question data based on the answer data modified in S1004. This process is similar to the process of updating the confidence level in S912. Next, in S1006, the inference unit 312 updates the estimation result according to the updated confidence level. Next, in S1007, the inference unit 312 transmits the updated estimation result to the smart glasses 130 via the communication unit 207. Upon receiving the estimation result, the smart glasses 130 update the estimation result display according to the received estimation result.

[0066] As described above, in the inference system of this embodiment, the inference device 100 determines whether the answer data should be user-based input data or sensor data, depending on the question data. Furthermore, if sensor data does not exist or if the sensor data indicates an abnormal value, the inference device 100 obtains user input. Thus, the inference device 100 of this embodiment determines whether sensor data can be used as answer data based on the question content. Therefore, there is no need to confirm the presence or absence of sensor data for each device being monitored and to construct a database with pre-set sensor data or input data based on the question data. This allows the inference device 100 to perform appropriate inferences on phenomena without incurring any cost.

[0067] As a first variant of the embodiment, the hardware structure of the inference system is not limited to the embodiment. As another example, the input management unit 311 and the inference unit 312 can be implemented in different information processing devices. In this case, the information processing device that functions as the input management unit 311 receives question data from the information processing device that functions as the inference unit 312, generates answer data corresponding to the question data, and sends the answer data to the information processing device that functions as the inference unit 312. In this way, at least a part of the functions and processing of the inference device 100 can also be achieved by, for example, making multiple CPUs, RAMs, ROMs, and savers cooperate. In addition, as another example, at least a part of the functions and processing of the inference device 100 can also be achieved using hardware circuits. In addition, the hardware that displays the inference results, etc. is not limited to the smart glasses 130, and as another example, it can also be the display unit of a PC used by the user.

[0068] As a second modification, the inference device 100 may refer to only one type of sensor data for inference. In this case, the process of specifying the sensor type is unnecessary, and information indicating the sensor type is not necessary in the word DB 304 .

[0069] As a third variant, in this embodiment, the diagnostic process information is output after the inference is completed, but the output timing of the diagnostic process information is not limited to the embodiment. As another example, it can also be appropriately output according to the user operation before the inference is completed. In this case, the inference device 100 outputs the obtained question and its answer as the diagnostic process information. In addition, when a sensor data change instruction is accepted, the inference device 100 can perform inference after updating the answer that has been obtained.

[0070] <Other implementation methods>

[0071] In addition, the present invention can also be implemented by executing the following processing. That is, the software (program) that implements the functions of the above-mentioned embodiments is supplied to a system or device via a network or various recording media. And, the computer (or CPU, MPU, etc.) of the system or device reads and executes the program.

[0072] As described above, according to each of the above embodiments, it is possible to provide an inference device that can appropriately infer a phenomenon without increasing costs.

[0073] While preferred embodiments of the present invention have been described in detail above, the present invention is not limited to such specific embodiments, and various modifications and changes are possible within the scope of the gist of the present invention as described in the claims.

Claims

1. An inference device for inferring countermeasures or factors for a failure or abnormal phenomenon occurring in a monitored device, characterized in that: The inference device has: a question obtaining unit for obtaining questions related to the phenomenon; A question determination unit, which determines whether the question is a qualitative question or a quantitative question; a sensor determination unit for determining, in the case of the quantitative question, whether sensor data from the device to be monitored can be acquired; a determination unit that determines the sensor data as data to be used for inference when the sensor data can be obtained, and determines user-input data as data to be used for inference when the sensor data cannot be obtained; an inference unit that, each time data for inference is determined, uses the data for inference to make an inference corresponding to the phenomenon and calculates the confidence level of the inference result; a recording unit that, after calculating a first confidence level for an inference result, records information indicating a possible error in the sensor data in a storage unit if the confidence level changes to a second confidence level smaller than the first confidence level due to inference using the sensor data; The types of sensor data include temperature or humidity.

2. The inference device according to claim 1, characterized in that The inference device further includes a first output unit configured to output information indicating that the error may exist.

3. The inference device according to claim 1 or 2, characterized in that: The inference device further comprises: a sensor data acquisition unit that acquires the sensor data when the sensor data can be acquired; and a data determination unit that determines whether the sensor data acquired by the sensor data acquisition unit is normal data based on a predetermined condition; The determination unit determines the sensor data as data for inference when the sensor data is the normal data, and determines the input data as data for inference when the sensor data is not the normal data.

4. The inference device according to claim 3, characterized in that The data determination unit determines whether the sensor data is normal data based on whether the value of the sensor data is within a predetermined allowable range.

5. The inference device according to claim 3, characterized in that The data determination unit determines whether the data is normal based on a time series change of sensor data detected before a processing time point.

6. The inference device according to claim 1 or 2, characterized in that: The inference device has: a second output unit that outputs data for inference determined by the determination unit; and a changing unit that changes the data used for inference when receiving a change instruction for the data used for inference, When the data used for inference is changed by the changing unit, the inference unit performs inference using the changed data.

Images