Computer program, method for generating learning model, information processing method, and information processing device

Abstract

Provided are a computer program, a method for generating a learning model, an information processing method, and an information processing device that can be expected to reduce the influence of noise or the like in state detection of substrate processing using a learning model.　A computer program according to an embodiment of the present invention causes a computer to execute processing for: acquiring sensor data related to substrate processing; inputting the acquired sensor data to a learning model to acquire sensor data outputted by the learning model; calculating individual errors for measurement values between the sensor data inputted to the learning model and the sensor data outputted by the learning model; assigning a weight to each of the plurality of calculated individual errors to calculate a total error; updating an internal parameter of the learning model on the basis of the calculated total error; and adjusting the weight to be assigned to each of the individual errors according to the convergence state of the individual error.

Description

Computer program, learning model generation method, information processing method, and information processing device

[0001] The present disclosure relates to a computer program, a method for generating a learning model, an information processing method, and an information processing device.

[0002] Patent Document 1 proposes an anomaly detection device in which an encoder compresses input data using compression parameters that match normal data, detects a first anomaly in the input data based on the logarithmic probability density on the prior distribution of the compressed data and the logarithmic density ratio between the compressed data distribution and the prior distribution, decodes the compressed data using decoding parameters that match normal data, and detects a second anomaly in the input data based on the difference between the decoded reconstructed data and the input data.

[0003] Japanese Patent Application Laid-Open No. 2018-49355

[0004] The present disclosure provides a computer program, a method for generating a learning model, an information processing method, and an information processing device that are expected to reduce the influence of noise, etc. in detecting the state of substrate processing using a learning model.

[0005] A computer program according to one embodiment is a computer program that causes a computer to execute a process of generating a learning model through machine learning that accepts as input sensor data including multiple measurement values ​​measured by multiple sensors related to substrate processing and outputs sensor data including a corresponding multiple measurement values, and causes the computer to execute the following processes: acquire the sensor data related to the substrate processing, input the acquired sensor data into a learning model, acquire the sensor data output by the learning model, calculate individual errors for each measurement value between the sensor data input to the learning model and the sensor data output by the learning model, weight each of the calculated individual errors to calculate a total error, update internal parameters of the learning model based on the calculated total error, and adjust the weights assigned to the individual errors depending on the convergence status of the individual errors.

[0006] According to the present disclosure, it is expected that the influence of noise and the like can be reduced in detecting the state of substrate processing using a learning model.

[0007] FIG. 1 is a schematic diagram for explaining an overview of an information processing system according to the present embodiment. FIG. 2 is a block diagram showing an example of a configuration of an information processing device according to the present embodiment. FIG. 3 is a schematic diagram showing an example of a configuration of a learning model according to the present embodiment. FIG. 4 is a flowchart showing an example of the procedure of a learning model generation process performed by the information processing device according to the present embodiment. FIG. 5 is a flowchart showing an example of the procedure of an anomaly detection process performed by the information processing device according to the present embodiment. FIG. 6 is a schematic diagram showing an example of a display of an anomaly detection result. FIG. 7 is a schematic diagram for explaining a data augmentation process performed by the information processing device according to the present embodiment. FIG. 8 is a schematic diagram for explaining preprocessing performed by the information processing device according to the present embodiment.

[0008] Specific examples of information processing systems according to embodiments of the present disclosure will be described below with reference to the drawings. Note that the present disclosure is not limited to these examples, but is defined by the claims, and is intended to include all modifications within the meaning and scope of the claims.

[0009] <System Overview> FIG. 1 is a schematic diagram illustrating an overview of an information processing system according to this embodiment. The information processing system according to this embodiment includes an information processing apparatus 1 and a substrate processing apparatus 3. The substrate processing apparatus 3 is an apparatus that performs various processes, such as CVD (Chemical Vapor Deposition), sputtering, or etching, on semiconductor substrates (wafers). In this embodiment, the substrate processing apparatus 3 includes multiple sensors 4, such as a sensor that measures the temperature and a sensor that measures the pressure within a chamber that accommodates the substrate to be processed. The substrate processing apparatus 3 repeatedly acquires (samples) data, such as temperature and pressure, measured by the multiple sensors 4 at predetermined intervals during substrate processing, such as etching, and transmits the data acquired from the multiple sensors 4 as sensor data to the information processing apparatus 1. Alternatively, the substrate processing apparatus 3 may store the data acquired from the multiple sensors 4 during substrate processing in a memory or the like, and then transmit the stored data to the information processing apparatus 1 after the substrate processing is completed. Note that information exchange between the substrate processing apparatus 3 and the information processing apparatus 1 may be performed not via wired or wireless communication but via a recording medium, such as a memory card or an optical disk.

[0010] The information processing device 1 is a device that performs processing to detect abnormalities and the like related to substrate processing performed in the substrate processing device 3, for example, based on sensor data obtained from a plurality of sensors 4 provided in the substrate processing device 3. The information processing device 1 may detect abnormalities and the like for a plurality of substrate processing devices 3. The information processing device 1 according to this embodiment collects sensor data from the substrate processing device 3, records it in a database, and performs machine learning using the recorded sensor data to generate a learning model 2 for detecting abnormalities and the like.

[0011] The learning model 2 according to this embodiment is a learning model that employs the configuration of a so-called autoencoder. An autoencoder is a learning model that combines an encoder that compresses input data and converts it into feature data with a reduced number of dimensions, and a decoder that restores this feature data to data with the original dimensions. Machine learning using an autoencoder is performed using a so-called unsupervised machine learning method that does not require training data in which correct values ​​are assigned to the input data. The information processing device 1 can perform machine learning on the learning model 2 of the autoencoder by updating internal parameters so that the input data and output data match.

[0012] Furthermore, when anomaly detection is performed using an autoencoder, normal data is used for machine learning of the autoencoder, and abnormal data is not required. An autoencoder that has undergone machine learning using normal data outputs normal data when normal data is input, but does not output abnormal data when abnormal data is input. The information processing device 1 inputs data that is unknown as to whether it is normal or abnormal into the autoencoder learning model 2, obtains the data output by the learning model 2, and can determine that the input data contains an abnormality if the difference between the input data and the output data exceeds a predetermined threshold. Note that the learning method of the autoencoder learning model and the method of anomaly detection using the autoencoder learning model are existing technologies, so detailed description will be omitted.

[0013] The information processing device 1 according to this embodiment records sensor data obtained from the substrate processing device 3 in a database, and performs machine learning of the autoencoder learning model 2 using the sensor data obtained when substrate processing was performed normally. The information processing device 1 then uses the machine-learned learning model 2 to detect an abnormality in the substrate processing performed by the substrate processing device 3. The information processing device 1 acquires sensor data measured by multiple sensors 4 in connection with the substrate processing from the substrate processing device 3, inputs the acquired sensor data to the learning model 2, and calculates the difference between the data output by the learning model 2 and the input sensor data. If the calculated difference exceeds a predetermined threshold, the information processing device 1 determines that there is an abnormality in the substrate processing, and notifies a user, such as an administrator of the information processing system, of the abnormality.

[0014] In an information processing system that performs anomaly detection using such an autoencoder learning model 2, the electrical signals output as measurement results of temperature, pressure, etc. from the multiple sensors 4 provided in the substrate processing apparatus 3 may contain noise unrelated to the substrate processing. In such cases, the information processing apparatus 1 may erroneously detect the noise contained in the acquired sensor data as an anomaly. In the information processing system according to this embodiment, machine learning of the learning model 2 is performed so as to suppress erroneous detection due to noise contained in the sensor data.

[0015] <Device Configuration> Fig. 2 is a block diagram showing an example configuration of an information processing device 1 according to this embodiment. The information processing device 1 according to this embodiment can be realized by installing a predetermined application program or the like in a general-purpose information processing device such as a personal computer or a server computer. The information processing device 1 according to this embodiment is configured to include a processing unit 11, a storage unit 12, a communication unit 13, a display unit 14, an operation unit 15, etc. Note that, in this embodiment, the processing will be described as being performed by a single information processing device 1, but the processing of the information processing device 1 may be distributed among multiple devices.

[0016] The processing unit 11 is configured using an arithmetic processing device such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit) or a quantum processor, a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The processing unit 11 reads and executes a program 12a stored in the storage unit 12, thereby performing various processes such as generating a learning model 2 based on sensor data obtained from the sensor 4 of the substrate processing apparatus 3, and detecting anomalies in the substrate processing using the generated learning model 2.

[0017] The storage unit 12 is configured using a large-capacity storage device such as a hard disk or an SSD (Solid State Drive). The storage unit 12 stores various programs executed by the processing unit 11 and various data required for the processing of the processing unit 11. In the present embodiment, the storage unit 12 stores a program 12a executed by the processing unit 11. The storage unit 12 also includes a sensor data storage unit 12b that stores sensor data acquired from the substrate processing apparatus 3, and a model information storage unit 12c that stores information related to the learning model 2 generated by the information processing apparatus 1.

[0018] In this embodiment, the program (computer program, program product) 12a is provided in a form recorded on a recording medium 99 such as a memory card or an optical disc, and the information processing device 1 reads the program 12a from the recording medium 99 and stores it in the storage unit 12. However, the program 12a may also be written to the storage unit 12, for example, during the manufacturing stage of the information processing device 1. Alternatively, the program 12a may be distributed by a remote server device or the like and acquired by the information processing device 1 via communication. For example, the program 12a may be read from the recording medium 99 by a writing device and written to the storage unit 12 of the information processing device 1. The program 12a may be provided in a form distributed via a network or in a form recorded on the recording medium 99.

[0019] The sensor data storage unit 12b stores the sensor data acquired from the substrate processing apparatus 3 together with information such as the date and time when the sensor data was acquired, identification information of the substrate processing apparatus 3, and identification information of the substrate processed by the substrate processing apparatus 3. The substrate processing apparatus 3 is equipped with a plurality of sensors 4, and the sensor data is obtained by repeatedly sampling signals output from the plurality of sensors 4 at a predetermined period during substrate processing. For example, if the substrate processing apparatus 3 is equipped with M sensors and sampling is performed N times during substrate processing, the sensor data acquired by the information processing apparatus 1 will contain M×N values. The sensor data storage unit 12b stores the sensor data including the M×N values ​​in association with information such as the date and time and identification information.

[0020] The model information storage unit 12c stores information about a learning model that has undergone machine learning. The information about the learning model may include, for example, information indicating the configuration of the learning model and information such as values ​​of internal parameters determined by machine learning. In the present embodiment, the model information storage unit 12c stores information about a learning model 2 of an autoencoder used for abnormality detection in the substrate processing apparatus 3. The learning model 2 according to the present embodiment is a learning model that has undergone machine learning to receive, for example, sensor data containing the above-mentioned M×N values ​​as input and output predicted values ​​of normal sensor data corresponding to the sensor data as M×N data. The information processing apparatus 1 generates the learning model 2 by performing machine learning using normal sensor data acquired from the substrate processing apparatus 3, and stores information about the generated learning model 2 in the model information storage unit 12c.

[0021] In this embodiment, information about the learning model is stored in the information processing device 1, and processing using the learning model is performed by the information processing device 1, but this is not limited to this. Information about the learning model may be stored in a device different from the information processing device 1, and this device may perform processing using the learning model, and the information processing device 1 may acquire the processing results from this device. Furthermore, machine learning processing of the learning model may be performed by the information processing device 1 or by a device different from the information processing device 1.

[0022] The communication unit 13 is connected to the substrate processing apparatus 3 via, for example, a communication line, and exchanges data with the substrate processing apparatus 3. In this embodiment, the communication unit 13 receives sensor data transmitted from the substrate processing apparatus 3 and provides the data to the processing unit 11. The communication unit 13 also transmits data such as control commands transmitted from the processing unit 11 to the substrate processing apparatus 3.

[0023] The display unit 14 is configured using a liquid crystal display or the like, and displays various images, characters, etc. based on the processing of the processing unit 11. The display unit 14 displays various information such as sensor data acquired from the substrate processing apparatus 3, information related to the generated learning model 2, and the operating status of the substrate processing apparatus 3 being controlled.

[0024] The operation unit 15 accepts user operations and notifies the processing unit 11 of the accepted operations. For example, the operation unit 15 accepts user operations using input devices such as mechanical buttons or a touch panel provided on the surface of the display unit 14. Furthermore, for example, the operation unit 15 may be input devices such as a mouse and a keyboard, and these input devices may be configured to be detachable from the information processing device 1.

[0025] The storage unit 12 may be an external storage device connected to the information processing device 1. The information processing device 1 may be a multi-computer including multiple computers, or may be a virtual machine virtually constructed by software. The information processing device 1 is not limited to the above configuration, and may not include, for example, the display unit 14 and the operation unit 15.

[0026] In the information processing device 1 according to this embodiment, the processing unit 11 reads and executes the program 12a stored in the storage unit 12, thereby realizing a sensor data acquisition unit 11a, a learning processing unit 11b, an abnormality detection unit 11c, a display processing unit 11d, etc. as software functional units in the processing unit 11. Note that in the figure, functional units related to processes such as collection of sensor data from the substrate processing device 3, generation of the learning model 2, and abnormality detection in the substrate processing device 3 are shown as functional units of the processing unit 11, and functional units related to processes other than these are not shown.

[0027] The sensor data acquiring unit 11a communicates with the substrate processing apparatus 3 via the communication unit 13, thereby acquiring sensor data including multiple measurement values ​​such as temperature and pressure measured by the multiple sensors 4 when the substrate processing apparatus 3 performs substrate processing. The substrate processing apparatus 3 stores the sensor data in a memory or the like by repeatedly sampling signals output by the multiple sensors 4 at a predetermined period during substrate processing, and transmits the stored sensor data to the information processing apparatus 1 at an appropriate timing, for example, after the substrate processing is completed. The sensor data acquiring unit 11a of the information processing apparatus 1 acquires the sensor data transmitted from the substrate processing apparatus 3 after the substrate processing is completed, and stores the acquired sensor data in the sensor data storage unit 12b with information such as the acquisition date and time and identification information of the substrate processing apparatus 3 attached to the acquired sensor data.

[0028] Alternatively, for example, while the substrate processing apparatus 3 is performing substrate processing, the sensor data acquiring unit 11a may communicate with the substrate processing apparatus 3 at a predetermined cycle to repeatedly acquire sensor data. This allows the sensor data acquiring unit 11a to acquire sensor data in real time as substrate processing is performed by the substrate processing apparatus 3. The cycle and timing at which the sensor data acquiring unit 11a acquires sensor data from the substrate processing apparatus 3 may be determined as appropriate by a designer of the information processing system, etc.

[0029] The learning processing unit 11b performs machine learning processing to generate a learning model 2 using the sensor data stored in the sensor data storage unit 12b. As described above, the learning model 2 has an autoencoder configuration, and receives sensor data including, for example, M×N values ​​as input and outputs data including M×N values ​​as well. The learning processing unit 11b uses sensor data from when substrate processing is normally performed among the sensor data stored in the sensor data storage unit 12b to update the internal parameters of the learning model 2 so that the sensor data input to the learning model 2 matches the data output by the learning model 2. In this way, the learning processing unit 11b generates a learning model 2 that predicts normal sensor data for input sensor data. The learning processing unit 11b stores information such as the internal parameters of the learning model 2 obtained by machine learning in the model information storage unit 12c.

[0030] Furthermore, in this embodiment, among the multiple sensors 4 provided in the substrate processing apparatus 3, sensors 4 that are difficult to predict using the learning model 2 are sensors whose sensor data contains noise unrelated to substrate processing and are considered to have little impact on subsequent abnormality detection, etc. In the machine learning process for generating the learning model 2, the learning processing unit 11b calculates a prediction error for each sensor 4 and reduces the learning importance of sensors 4 whose errors are difficult to converge, thereby proceeding with machine learning so as to prevent the learning model 2 from outputting sensor data containing noise as correct sensor data. Furthermore, the learning processing unit 11b stores information about the sensors 4 that have been determined to be difficult to predict and have had their learning importance reduced, together with information such as the internal parameters of the learning model 2, in the model information storage unit 12c.

[0031] The abnormality detection unit 11c performs processing to detect abnormalities related to substrate processing based on sensor data acquired from the substrate processing apparatus 3, using a learning model 2 previously generated by machine learning. The abnormality detection unit 11c reconstructs and uses the learning model 2 based on information stored in the model information storage unit 12c. The abnormality detection unit 11c acquires sensor data measured by multiple sensors 4, for example, while the substrate processing apparatus 3 is performing substrate processing or after the substrate processing is completed. The abnormality detection unit 11c inputs the sensor data acquired from the substrate processing apparatus 3 to the learning model 2, and acquires data output by the learning model 2 in response to the sensor data. The abnormality detection unit 11c determines the presence or absence of an abnormality related to substrate processing based on a difference between the data input to the learning model 2 and the data output by the learning model 2.

[0032] The anomaly detection unit 11c may individually perform anomaly detection for each of the multiple sensors 4 included in the substrate processing apparatus 3. For example, the anomaly detection unit 11c calculates an error such as a mean square error for each sensor 4 between the input data to the learning model 2 and the output data of the learning model 2, and for a sensor 4 for which the calculated error exceeds a threshold, it can determine that an anomaly has occurred in the substrate processing event measured by that sensor 4. In addition, for sensors 4 that have been determined to be difficult to predict in the machine learning of the learning model 2 and have had their learning importance lowered, the anomaly detection unit 11c does not need to perform processes such as error calculation and comparison with a threshold on the sensors 4, as they are excluded from anomaly detection.

[0033] The display processing unit 11d performs processing to display various characters, images, and the like on the display unit 14. In the present embodiment, the display processing unit 11d graphs sensor data acquired from, for example, the substrate processing apparatus 3 and displays the graph on the display unit 14. The display processing unit 11d also displays the progress of the machine learning processing performed by the learning processing unit 11b, for example, a graph showing changes in the prediction error of the learning model 2, on the display unit 14. The display processing unit 11d also displays information on the results of abnormality detection in the substrate processing using the learning model 2, such as whether or not there is an abnormality in the substrate processing, and, if there is an abnormality, which sensor 4 detected the abnormality. The display processing unit 11d may also display various other information on the display unit 14.

[0034] <Learning Model Generation Process> Figure 3 is a schematic diagram showing an example of the configuration of the learning model 2 according to this embodiment. The learning model 2 according to this embodiment is a learning model configured as an autoencoder including an encoder and a decoder (not shown). The encoder and decoder included in the learning model 2 may be configured as, for example, a neural network. The encoder of the learning model 2 converts input data into features with a small number of dimensions, and the decoder inversely converts these features into data with the original dimensions. The learning model 2 receives as input sensor data measured by a plurality of sensors 4 included in the substrate processing apparatus 3, and performs machine learning to output predicted values ​​of normal sensor data for the input sensor data.

[0035] The information processing device 1 acquires sensor data when the substrate processing device 3 performs substrate processing, and stores and accumulates the acquired normal sensor data in the sensor data storage unit 12b. The sensor data stored in the sensor data storage unit 12b is accompanied by information indicating whether the substrate processing was performed normally, for example, by a user who has confirmed the results of the substrate processing. In the present embodiment, whether the sensor data is normal is determined by the user, but this is not limited to this. For example, the information processing device 1 may perform processing such as measuring the results of the substrate processing with a measuring device and determining whether the measurement value obtained by the measurement is within a predetermined range. Based on this determination result, the information processing device 1 may associate normal or abnormal information with the sensor data and store it in the sensor data storage unit 12b. Based on this information, the information processing device 1 can extract normal sensor data from the sensor data stored in the sensor data storage unit 12b. The information processing device 1 performs machine learning processing to generate a learning model 2 using sensor data from the sensor data storage unit 12b when the substrate processing was performed normally.

[0036] In this figure, the sensor data that is input to and output from the learning model 2 is shown as a waveform. However, in reality, for example, if the substrate processing apparatus 3 is equipped with M sensors and N samplings are performed during substrate processing, both the input and output data to the learning model 2 can be an M×N matrix. The information processing apparatus 1 reads out the sensor data stored in the sensor data storage unit 12b, and performs preprocessing such as normalization or standardization on the read sensor data as necessary to obtain learning data for machine learning of the learning model 2.

[0037] The information processing device 1 inputs learning sensor data read from the sensor data storage unit 12b to the learning model 2, whose internal parameters have been set to appropriate initial values ​​in advance, and acquires data output by the learning model 2. As described above, the sensor data input to the learning model 2 is M×N matrix data, and the data output by the learning model 2 is also M×N matrix data. Based on the input data to the learning model 2 and the output data of the learning model 2, the information processing device 1 according to this embodiment calculates the squared error for each component of the M×N matrix using the following equation (1) as the individual error L t current By this calculation, the information processing device 1 obtains M×N individual errors. t input are the matrix elements of the input data, and x t pred are the elements of the matrix of the corresponding output data. In addition, in equation (1) and the following equations, the superscript t indicates the number of iterations in machine learning, that is, the number of epochs.

[0038]

[0039] Next, the information processing device 1 calculates each individual error L t current Regarding the exponential moving average error L t ema is calculated based on the following equation (2). By this calculation, the information processing device 1 calculates M×N exponentially smoothed moving average errors L t emaHere, in equation (2) and the following equations, the coefficient α is a hyperparameter determined by the designer or the like of the information processing system according to this embodiment. The information processing device 1 calculates the current exponentially smoothed moving average error based on the previous exponentially smoothed moving average error in the machine learning iteration and the current individual error. For this purpose, the information processing device 1 stores the calculation result of the current exponentially smoothed moving average error in the storage unit 12 at least until the next exponentially smoothed moving average error is calculated.

[0040]

[0041] Next, the information processing device 1 calculates a machine learning weight coef. for each component of the M×N matrix of sensor data based on the following equation (3). Through this calculation, the information processing device 1 obtains M×N weight coef.s. In (3), the coefficient β is a hyperparameter determined by the designer or the like of the information processing system according to this embodiment. For components of the sensor 4 whose errors are difficult to converge by repeated machine learning, the value of the weight coef. becomes smaller. The information processing device 1 can assign weights such that the smaller the individual error at a certain point in time is compared to the previous individual error, the larger the weight assigned to this individual error.

[0042]

[0043] Next, the information processing device 1 calculates the M×N individual errors L t current and the corresponding M × N weights coef. are used to calculate the overall error L t is calculated based on the following formula (4): t is M × N individual errors L t current The sum of the M×N values ​​obtained by multiplying them by the weights corresponding to the respective values ​​is calculated to obtain a single numerical value.

[0044]

[0045] The information processing device 1 calculates the total error L t Using this total error L t Machine learning of the learning model 2 is performed by updating the internal parameters of the learning model 2 so that is small. The information processing device 1 can update the internal parameters of the learning model 2 by, for example, the backpropagation method. The process of updating the internal parameters of the learning model 2 by the backpropagation method is an existing technology, so a detailed description will be omitted. The total error L calculated based on the above equations (1) to (4) t The weight coef. of the value relating to the sensor 4 for which the error is difficult to converge, such as a sensor 4 with large noise, becomes smaller, and the importance or influence of such a sensor 4 in machine learning becomes a value that is reduced. t By performing machine learning based on this, the learning model 2 does not learn the characteristics of a sensor 4 that has a lot of noise and is difficult to predict, but learns the characteristics of a sensor 4 that has little noise and is easy to predict.

[0046] 4 is a flowchart showing an example of the procedure of a learning model generation process performed by the information processing device 1 according to this embodiment. The processing unit 11 of the information processing device 1 according to this embodiment initializes the value of a variable t used in this process to 1 (step S1). The variable t is a variable for counting the number of iterations of machine learning, and is realized using, for example, a register in the processing unit 11 or a storage area in the storage unit 12. The learning processing unit 11b of the processing unit 11 acquires, from the sensor data stored in the sensor data storage unit 12b of the storage unit 12, sensor data measured by the sensor 4 when substrate processing was performed normally, as sensor data to be used for machine learning (step S2).

[0047] The learning processing unit 11b inputs the sensor data acquired in step S2 to the learning model 2 (step S3). The learning processing unit 11b acquires data output by the learning model 2 in response to the sensor data input in step S3 (step S4). The learning processing unit 11b calculates individual errors for each component of the sensor data input to the learning model 2 in step S3 and the output data of the learning model 2 acquired in step S4, based on the above-mentioned equation (1) (step S5). The learning processing unit 11b calculates weights for each component of the sensor data based on the above-mentioned equations (2) and (3) using the individual errors calculated in step S5 and the previous exponentially smoothed moving average error (step S6). The learning processing unit 11b calculates a total error based on the above-mentioned equation (4) using the individual errors calculated in step S5 and the weights calculated in step S6 (step S7). Based on the total error calculated in step S7, the learning processing unit 11b updates the internal parameters of the learning model 2 using a method such as the backpropagation method so as to reduce the total error (step S8).

[0048] The learning processing unit 11b determines whether to terminate the machine learning process based on conditions such as whether the number of machine learning iterations has reached a predetermined number or whether the total error calculated in step S7 has fallen below a predetermined target value (step S9). If the machine learning process is not to be terminated (S9: NO), the learning processing unit 11b adds 1 to the variable t (step S10) and returns to step S3. If the machine learning process is to be terminated (S9: YES), the learning processing unit 11b stores information such as the final internal parameters of the learning model 2 and the weights of each component of the sensor data obtained during the machine learning process in the model information storage unit 12c (step S11), and terminates the process. Note that the weights of each component stored in step S11 may be only representative values, such as the minimum or average value, of the weights obtained through multiple iterations of machine learning.

[0049] <Abnormality Detection Processing> The information processing device 1 according to this embodiment uses the learning model 2 generated by the above-described machine learning to perform processing to detect abnormalities related to substrate processing by the substrate processing device 3. The information processing device 1 according to this embodiment acquires sensor data including various measurement values ​​measured by the multiple sensors 4 during the substrate processing process, for example, while the substrate processing device 3 is performing substrate processing or after the substrate processing has been completed. The information processing device 1 inputs the acquired sensor data (or data obtained by performing appropriate preprocessing on this sensor data) to the trained learning model 2 and acquires data output by the learning model 2 in response to this.

[0050] The information processing device 1 compares each component of the data input to the learning model 2 with the data output by the learning model 2, and if the difference (error) between each component exceeds a predetermined threshold, it determines that there is an abnormality in the sensor 4 related to the component whose difference exceeds the threshold. The threshold used for abnormality detection may be, for example, a common value for multiple components, or, for example, an individual value may be set for each component. These thresholds are predetermined by a designer of the information processing system according to this embodiment and are pre-stored in the storage unit 12 of the information processing device 1. Furthermore, the difference between each component may be, for example, a differential value between two input and output values ​​or its absolute value, or may be, for example, an individual error according to the above formula (1), or a value obtained by a calculation other than these.

[0051] The information processing device 1 may determine the presence or absence of an abnormality for each sensor 4, rather than for each component of input / output data. For example, when sensor data is obtained by performing sampling N times for each of M sensors 4, the input / output data to the learning model 2 becomes an M×N matrix. When determining the presence or absence of an abnormality for each component, the information processing device 1 can calculate the M×N differences and compare them with a threshold. When determining the presence or absence of an abnormality for each sensor 4, the information processing device 1 treats the N values ​​for each sensor 4 as an N-dimensional vector, calculates the similarity between the two input / output N-dimensional vectors, and can determine that a sensor 4 whose similarity is less than a threshold is abnormal.

[0052] As described above, the information processing device 1 according to the present embodiment sets a small weight for a sensor 4 with high noise and low error convergence during machine learning to prevent learning of its characteristics. Therefore, the trained learning model 2 may have low prediction accuracy for a sensor 4 with a small weight, potentially preventing accurate anomaly detection. Therefore, the information processing device 1 stores the weights calculated during machine learning along with information such as the internal parameters of the learning model 2 obtained through machine learning, and excludes sensors 4 with weights lower than a predetermined threshold from anomaly detection. The information processing device 1 may not store all weights obtained during machine learning, but may instead store, for example, the minimum or average weight for each sensor 4. Alternatively, the information processing device 1 may store, for example, identification information of sensors 4 to be excluded from anomaly detection based on the weights, rather than storing the weights.

[0053] The information processing device 1 displays information about the results of the abnormality detection on the display unit 14. The information processing device 1 displays a list of, for example, identification information for a plurality of sensors 4 provided in the substrate processing device 3, in association with the presence or absence of an abnormality for each sensor 4. At this time, the information processing device 1 may also display information about which sensors 4 were excluded from the abnormality detection target. Furthermore, the information processing device 1 may display more detailed information about the sensors 4 in which an abnormality was detected, for example by displaying sensor data in a graph.

[0054] 5 is a flowchart showing an example of the procedure of an abnormality detection process performed by the information processing device 1 according to this embodiment. The sensor data acquisition unit 11a of the processing unit 11 of the information processing device 1 according to this embodiment acquires sensor data including measurement values ​​of the plurality of sensors 4 provided in the substrate processing device 3 by communicating with the substrate processing device 3 via the communication unit 13 (step S31).

[0055] The anomaly detection unit 11c of the processing unit 11 constructs a learning model 2 based on the information stored in the model information storage unit 12c, and inputs the sensor data acquired in step S31 to the learning model 2 (step S32). The anomaly detection unit 11c acquires data output by the learning model 2 in response to the sensor data input in step S32 (step S33). The anomaly detection unit 11c compares the sensor data input to the learning model 2 in step S32 with the output data of the learning model 2 acquired in step S33, and calculates the error of each component of the input / output data based on, for example, the above-mentioned formula (1) (step S34).

[0056] The anomaly detection unit 11c acquires information on the weight of each component stored in the model information storage unit 12c when machine learning of the learning model 2 was performed, and excludes sensors 4 including components whose weights are lower than a predetermined threshold from the targets for anomaly detection (step S35). For sensors 4 not excluded in step S35, the anomaly detection unit 11c determines whether the error calculated in step S34 exceeds a predetermined threshold (step S36). If the error exceeds the threshold (S36: YES), the anomaly detection unit 11c notifies that an anomaly has been detected for the sensor 4 whose error exceeds the threshold (step S37), and terminates the process. If the error does not exceed the threshold (S36: NO), the anomaly detection unit 11c terminates the process.

[0057] 6 is a schematic diagram showing an example of a display of anomaly detection results. As a result of substrate processing anomaly detection based on sensor data acquired from the substrate processing apparatus 3, the information processing apparatus 1 according to this embodiment displays information about the sensors 4 targeted for anomaly detection on the display unit 14 in association with the degree of anomaly (error) of each sensor. In the example shown in FIG. 6, the information processing apparatus 1 displays information about the sensors 4 targeted for anomaly detection in association with the degree of anomaly, by arranging the multiple sensors 4 included in the substrate processing apparatus 3 in descending order of the degree of anomaly (in descending order of the largest error) based on the errors of each sensor 4 obtained by the anomaly detection process. In this example, the information processing apparatus 1 displays the errors of each sensor 4 as a bar graph at the top of the screen, with the sensors 4 with the largest errors arranged from left to right.

[0058] The display of the abnormality detection results by the information processing device 1 is not limited to the example shown in FIG. 6 , and various display modes may be employed. The information processing device 1 may display the abnormality level (error) of each sensor 4 using various graphs, such as line graphs or pie charts, instead of bar graphs. The information processing device 1 may display information about each sensor 4 in a table format instead of a graph. In this case, for example, the information about the sensors 4 may be displayed in a table format in descending order of the abnormality level. Furthermore, instead of displaying information about the sensors 4 in descending order of the abnormality level, the information processing device 1 may display the information in an appropriate order, such as in order of the identification number of the sensors 4, and highlight the information about the sensors 4 with the highest abnormality level, thereby displaying information about the sensors 4 targeted for abnormality detection in association with the abnormality level. For example, highlighting may be achieved by changing the color, size, or font of the text displaying the information depending on the abnormality level, or by flashing the information about the sensors 4 with the highest abnormality level.

[0059] Furthermore, the information processing device 1 excludes sensors 4 for which the weight of individual errors was set low during machine learning from those to be displayed, or displays them in a manner that makes it clear that they are not subject to anomaly detection. For example, the information processing device 1 can distinguishably display information about the target sensors 4 from information about the non-target sensors 4 by providing a difference in display mode, such as displaying information about the target sensors 4 for anomaly detection and information about the non-target sensors 4 in different colors or sizes. Furthermore, for example, the information processing device 1 may display information about the non-target sensors 4 with an icon or the like indicating that the information is not subject to anomaly detection.

[0060] Also in this figure, the information processing device 1 displays, at the bottom of the screen, a graph of the time-series sensor data used for anomaly detection for the sensor 4 with the largest error (sensor K in the figure) and a graph of the time-series error calculated for this time-series sensor data. Note that in this example, the information processing device 1 displays a graph of the sensor data and error for the sensor 4 with the largest error, but this is not limited to this. For example, the user can perform an operation to select the sensor name of multiple sensors 4 arranged in order of largest error at the top of the screen, and the information processing device 1, upon accepting this operation, displays a graph of the sensor data and error for the sensor 4 selected by the user at the bottom of the screen.

[0061] <Additional Function 1: Data Expansion> The information processing device 1 performs machine learning to generate a learning model 2 using a plurality of pieces of sensor data acquired from the substrate processing device 3 and stored in the sensor data storage unit 12b. The more learning data used in the machine learning, the higher the predictive accuracy of the generated learning model 2 can be expected. Therefore, the information processing device 1 according to this embodiment can perform a process of increasing the learning data, that is, so-called data expansion, by generating new sensor data that can be used for machine learning based on the sensor data acquired from the substrate processing device 3.

[0062] 7 is a schematic diagram for explaining the data extension process performed by the information processing device 1 according to this embodiment. The information processing device 1 according to this embodiment uses the time-series measurement values ​​of each sensor 4 included in the sensor data previously acquired from the substrate processing device 3 as original data, and generates one or more pieces of extended data by performing appropriate calculations on this original data. The information processing device 1 determines, as the data extension range, a range obtained by extending the original data by a predetermined amount in each direction of the vertical axis (measurement value) and the horizontal axis (time). The upper part of FIG. 7 shows a step waveform as an example of original data, and the middle part shows an example of an extension range for this waveform.

[0063] In this embodiment, the information processing device 1 determines an extension range for the original data based on information about the resolution of the sensor 4 that output the original data (for example, the vertical axis is determined by the minimum value that the sensor 4 can measure, and the horizontal axis is determined by the sampling period). For example, the information processing device 1 can determine the extension range as a range in which the original data can be moved by plus or minus the resolution on the vertical axis and horizontal axis, respectively.

[0064] The information processing device 1 generates one or more pieces of extended data by appropriately performing operations such as movement, enlargement, or reduction on the original data so that the data fits within the determined extension range. The lower part of Figure 7 shows waveforms of three pieces of extended data generated to fit within the extension range. The extended data shown on the left is the original data reduced in the vertical axis direction. The extended data shown in the center is the original data expanded in the vertical axis direction. The extended data shown on the right is the original data translated in the vertical axis direction. The information processing device 1 randomly selects, for example, from multiple types of operations such as predetermined movement, enlargement, or reduction, and performs the selected operation on the original data to extend the data within the extension range.

[0065] The information processing device 1 generates one or more extended sensor data for one piece of sensor data (normal sensor data) stored in the sensor data storage unit 12b, and can more than double the amount of sensor data that can be used for machine learning of the learning model 2. By performing machine learning using a larger amount of learning data, the information processing device 1 is expected to generate a learning model 2 with higher prediction accuracy. Furthermore, by determining the extension range for data extension based on the resolution of the sensor 4, it is expected that erroneous detection of an abnormality due to the resolution of the sensor 4 can be prevented.

[0066] <Additional Function 2: Preprocessing> The information processing apparatus 1 may perform appropriate preprocessing on the sensor data acquired from the substrate processing apparatus 3, and input the preprocessed sensor data to the learning model 2. For example, standardization, normalization, or the like may be employed for the preprocessing. Standardization is a process of converting a plurality of values ​​based on their average value and standard deviation so that the average value is 0 and the variance is 1. Normalization is a process of converting a plurality of values ​​based on their minimum and maximum values ​​so that the values ​​fall within a range from 0 to 1. Note that processes such as standardization and normalization are existing technologies, and therefore detailed explanations thereof will be omitted.

[0067] FIG. 8 is a schematic diagram for explaining preprocessing performed by the information processing device 1 according to this embodiment. In this embodiment, the information processing device 1 performs preprocessing using summary values ​​of sensor data, and inputs the preprocessed sensor data to the learning model 2. The summary values ​​of the sensor data are the mean value and standard deviation when standardization is performed as preprocessing, and are the minimum value and maximum value when normalization is performed as preprocessing. Note that the information processing device 1 may perform preprocessing other than standardization and normalization, and in this case, the summary values ​​of the sensor data may be various values ​​required for preprocessing. Below, a case where standardization is performed as preprocessing will be described.

[0068] The information processing device 1 can obtain sensor data including M×N values ​​by, for example, performing sampling N times on M sensors 4. The information processing device 1 can obtain two summary values ​​by calculating the average value and standard deviation of the N values ​​for each sensor 4, and can obtain 2×M summary values ​​from one piece of sensor data.

[0069] The information processing device 1 according to this embodiment has a summary value prediction model 5 generated in advance by machine learning (i.e., information related to the prediction model 5 is stored in the model information storage unit 12c). The summary value prediction model 5 is a learning model that receives a summary value as input and has undergone machine learning to predict a corresponding normal summary value. In this embodiment, the information processing device 1 uses a model that performs dimensionality reduction by principal component analysis as the summary value prediction model 5, converts 2×M summary values ​​into information on a smaller number of principal components, and then inversely converts the converted principal component information back into 2×M summary values. In this way, the information processing device 1 can obtain a predicted normal summary value for sensor data acquired from the substrate processing device 3 based on the summary value of the sensor data.

[0070] The information processing device 1 preprocesses the sensor data including the original M×N values ​​using predicted values ​​of the 2×M summary values ​​obtained from the prediction model 5. In this example, the sensor data includes values ​​for M sensors 4, and N time-series values ​​for each sensor 4. The information processing device 1 predicts two summary values, the mean value and standard deviation, for each sensor 4, and standardizes the N time-series values ​​using these predicted mean values ​​and standard deviation values. The information processing device 1 can similarly perform standardization for each of the M sensors 4, thereby standardizing all of the M×N values ​​included in the sensor data.

[0071] The prediction model 5 used in preprocessing is generated by the information processing device 1 before machine learning to generate the learning model 2 is performed. The information processing device 1 acquires normal sensor data to be used in machine learning of the learning model 2 and calculates a summary value for each sensor 4 for the acquired sensor data. The information processing device 1 reduces the dimensions of the calculated summary values ​​using principal component analysis or the like, and generates a prediction model 5 that restores the reduced-dimension information to its original dimensions. Note that principal component analysis is an existing technology, so a detailed description will be omitted. Furthermore, the prediction model 5 is not limited to one based on principal component analysis, and any learning model that can reduce the number of dimensions of input data and restore it to its original dimensions may be adopted. The prediction model 5 may be any learning model that can predict a normal summary value for an input summary value.

[0072] The information processing device 1 also performs preprocessing using a pre-generated prediction model 5 both when generating a learning model 2 by machine learning and when performing anomaly detection using the trained learning model 2. The same prediction model 5 (having the same internal parameters determined by machine learning) is used in both cases. The information processing device 1 generates the prediction model 5 in advance and stores information such as the internal parameters of the prediction model 5 in the model information storage unit 12c.

[0073] The information processing device 1 according to this embodiment predicts a normal summary value from the summary value of the sensor data using the prediction model 5, and performs preprocessing such as standardization or normalization using the predicted normal summary value. As a result, the information processing device 1 reduces or nullifies changes in values ​​caused by factors such as individual differences or changes over time in the substrate processing device 3 through preprocessing, and is expected to improve the prediction accuracy of the learning model 2.

[0074] <Summary> In the information processing system according to the present embodiment configured as described above, the information processing device 1 receives as input sensor data including multiple measurement values ​​measured by multiple sensors 4 provided in the substrate processing device 3, and generates a learning model 2 through machine learning. The learning model 2 outputs sensor data including multiple corresponding normal measurement values. The information processing device 1 acquires normal sensor data related to substrate processing from the sensor data stored in the sensor data storage unit 12b, inputs the acquired sensor data to the learning model 2, and acquires sensor data output by the learning model 2. The information processing device 1 calculates individual errors for each measurement value included in the sensor data input to the learning model 2 and the sensor data output by the learning model 2, and calculates a total error by weighting the calculated individual errors. The information processing device 1 performs machine learning by updating the internal parameters of the learning model 2 based on the calculated total error, thereby generating the learning model 2. The information processing device 1 also adjusts the weights assigned to the individual errors depending on the convergence status of the individual errors. As a result, the information processing system according to the present embodiment can generate the learning model 2 by performing machine learning that takes into account, for example, sensor data with high noise levels and sensor data with low noise levels. This is expected to reduce the influence of noise and other factors when detecting the state of substrate processing using the generated learning model 2.

[0075] Furthermore, in the information processing system according to this embodiment, the information processing device 1 calculates weights based on the above-described formulas (1) to (3). As a result, the information processing device 1 adjusts the weights assigned to each individual error so that the smaller the individual error at a certain point in time is compared to the previous individual error, the greater the weight assigned to this individual error. In other words, the information processing device 1 adjusts the weights assigned to each individual error so that the weight assigned to an individual error that is difficult to converge is reduced. As a result, the information processing system according to this embodiment can be expected to generate a learning model 2 by reducing the importance of machine learning for a sensor 4 that is noisy and difficult to converge its individual error, for example.

[0076] Furthermore, in the information processing system according to this embodiment, the learning model 2 generated by the information processing device 1 has an autoencoder configuration. As a result, the information processing system according to this embodiment can perform machine learning to generate the learning model 2 using learning data that does not have a correct answer label, that is, so-called unsupervised machine learning. By performing machine learning using normal sensor data, the information processing device 1 is expected to generate a learning model 2 that predicts normal sensor data.

[0077] Furthermore, in the information processing system according to this embodiment, the information processing device 1 acquires sensor data from the substrate processing device 3, the sensor data including multiple sensor values ​​measured by multiple sensors 4 related to substrate processing. The information processing device 1 inputs the acquired sensor data to a learning model 2 generated in advance by the above-described machine learning, and acquires the sensor data output by the learning model 2. The information processing device 1 detects a state, such as an abnormality, related to the substrate processing based on a comparison between the sensor data input to the learning model 2 and the sensor data output by the learning model 2. As a result, the information processing system according to this embodiment is expected to reduce the influence of noise, etc., in detecting the state of substrate processing using the learning model 2.

[0078] Furthermore, in the information processing system according to this embodiment, the information processing device 1 stores, in the model information storage unit 12c, information related to the weights for each measurement value of the sensor data calculated during machine learning to generate the learning model 2. During state detection, the information processing device 1 determines sensors to be excluded from state detection based on the stored weights. This allows the information processing system according to this embodiment to exclude from state detection sensors 4 whose learning importance has been reduced because individual errors are difficult to converge during machine learning, and is expected to prevent erroneous detection of the state of a sensor 4 whose prediction accuracy by the learning model 2 is low.

[0079] The embodiments disclosed herein are to be considered as illustrative in all respects and not restrictive. The scope of the present disclosure is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims.

[0080] The matters described in each embodiment can be combined with each other. Furthermore, the independent claims and dependent claims described in the claims can be combined with each other in any and all combinations, regardless of the reference format. Furthermore, the claims use a format in which a claim references two or more other claims (multiple claim format), but this is not limited to this. A multiple claim (multi-multi claim) that references at least one other multiple claim may also be used.

[0081] REFERENCE SIGNS LIST 1 Information processing device (computer) 2 Learning model 3 Substrate processing device 4 Sensor 5 Prediction model 11 Processing unit 11a Sensor data acquisition unit 11b Learning processing unit 11c Abnormality detection unit 11d Display processing unit 12 Storage unit 12a Program (computer program) 12b Sensor data storage unit 12c Model information storage unit 13 Communication unit 14 Display unit 15 Operation unit

Claims

1. A computer program that causes a computer to execute a process of generating a learning model through machine learning that receives as input sensor data including multiple measurement values ​​measured by multiple sensors related to substrate processing and outputs sensor data including the corresponding multiple measurement values, the computer program causing the computer to execute the following processes: acquire the sensor data related to the substrate processing; input the acquired sensor data into a learning model to acquire the sensor data output by the learning model; calculate individual errors for each measurement value between the sensor data input to the learning model and the sensor data output by the learning model; calculate an overall error by weighting each of the calculated individual errors; update internal parameters of the learning model based on the calculated overall error; and adjust the weights assigned to the individual errors depending on the convergence status of the individual errors.

2. The computer program according to claim 1, wherein the weight assigned to each individual error is adjusted so that the smaller the individual error at a certain point in time is compared with the individual error at a point in time prior to said certain point in time, the greater the weight assigned to said individual error at said certain point in time.

3. The computer program of claim 1, wherein the learning model is an autoencoder.

4. The computer program according to claim 1, which generates extended sensor data by varying the measurement values ​​of the acquired sensor data within a predetermined range based on the acquired sensor data, and performs machine learning of the learning model using the generated extended sensor data as sensor data.

5. The computer program according to claim 1, further comprising: calculating a summary value of acquired sensor data; generating a prediction model that predicts the summary value of the sensor data based on the calculated summary value; preprocessing the sensor data using the summary value predicted by the generated prediction model; and inputting the preprocessed sensor data into the learning model.

6. A computer program that causes a computer to execute the following processes: acquire sensor data including multiple measurement values ​​taken by multiple sensors related to substrate processing; input the acquired sensor data into a learning model that has been machine-learned to accept the sensor data as input and output corresponding sensor data, thereby acquiring the sensor data output by the learning model; and detect a state related to the substrate processing based on a comparison between the sensor data input to the learning model and the sensor data output by the learning model, wherein the machine learning of the learning model is performed by: inputting sensor data into the learning model to acquire the sensor data output by the learning model; calculating individual errors for each measurement value between the sensor data input to the learning model and the sensor data output by the learning model; calculating a total error by weighting each of the calculated individual errors; updating internal parameters of the learning model based on the calculated total error; and adjusting the weight assigned to the individual error depending on the convergence status of the individual errors.

7. The computer program according to claim 6, further comprising: acquiring information relating to weights for each measurement value of the sensor data when the machine learning is performed; and determining sensors to be excluded from state detection based on the acquired weights.

8. The computer program according to claim 7, which calculates the degree of abnormality for each sensor based on a comparison between the sensor data input to the learning model and the sensor data output by the learning model, and outputs information about one or more sensors that are the subject of state detection in association with the degree of abnormality of the sensor.

9. The computer program according to claim 8, which outputs information relating to a plurality of sensors in order according to the degree of abnormality.

10. The computer program according to claim 7, which outputs information relating to sensors that are targets of status detection and information relating to sensors that are not targets of status detection in a manner that allows discrimination.

11. The computer program according to claim 6, further comprising: calculating a summary value of acquired sensor data; predicting a summary value of the sensor data based on the calculated summary value using a prediction model that has been machine-learned to accept the summary value of the sensor data as input and predict the summary value of the sensor data; preprocessing the acquired sensor data using the predicted summary value; and inputting the preprocessed sensor data into the learning model.

12. A method for generating a learning model in which an information processing device receives as input sensor data including multiple measurement values ​​measured by multiple sensors related to substrate processing, and generates a learning model through machine learning to output sensor data including corresponding multiple measurement values, the method comprising: acquiring the sensor data related to the substrate processing; inputting the acquired sensor data into a learning model to acquire sensor data output by the learning model; calculating individual errors for each measurement value between the sensor data input to the learning model and the sensor data output by the learning model; calculating a total error by weighting each of the calculated multiple individual errors; updating internal parameters of the learning model based on the calculated total error; and adjusting the weights assigned to the individual errors depending on the convergence status of the individual errors.

13. An information processing method in which an information processing device causes a computer to execute the following processes: acquire sensor data including multiple measurement values ​​taken by multiple sensors related to substrate processing; input the acquired sensor data into a learning model that has been machine-learned to accept the sensor data as input and output corresponding sensor data, thereby acquiring the sensor data output by the learning model; and detect a state related to the substrate processing based on a comparison between the sensor data input to the learning model and the sensor data output by the learning model, wherein the machine learning of the learning model is performed by inputting sensor data to the learning model to acquire the sensor data output by the learning model, calculating individual errors for each measurement value between the sensor data input to the learning model and the sensor data output by the learning model, calculating a total error by weighting each of the calculated multiple individual errors, updating internal parameters of the learning model based on the calculated total error, and adjusting the weight assigned to the individual error depending on the convergence status of the individual errors.

14. An information processing device comprising a processing unit that receives as input sensor data including multiple measurement values ​​measured by multiple sensors related to substrate processing, and generates a learning model through machine learning to output sensor data including corresponding multiple measurement values, wherein the processing unit: acquires the sensor data related to the substrate processing; inputs the acquired sensor data into a learning model to acquire sensor data output by the learning model; calculates individual errors for each measurement value between the sensor data input to the learning model and the sensor data output by the learning model; calculates a total error by weighting each of the calculated multiple individual errors; updates internal parameters of the learning model based on the calculated total error; and adjusts the weights assigned to the individual errors depending on the convergence status of the individual errors.

15. An information processing apparatus comprising a processing unit, wherein the processing unit causes a computer to execute the following processes: acquire sensor data including a plurality of measurement values ​​measured by a plurality of sensors related to substrate processing; input the acquired sensor data into a learning model that has been machine-learned to accept the sensor data as input and output corresponding sensor data, thereby acquiring the sensor data output by the learning model; and detect a state related to the substrate processing based on a comparison between the sensor data input to the learning model and the sensor data output by the learning model; wherein the machine learning of the learning model is performed by inputting sensor data to the learning model to acquire the sensor data output by the learning model; calculating individual errors for each measurement value between the sensor data input to the learning model and the sensor data output by the learning model; calculating a total error by weighting each of the calculated individual errors; updating internal parameters of the learning model based on the calculated total error; and adjusting the weight assigned to the individual error depending on the convergence status of the individual errors.

Images