An intelligent calibration method, device and system for metrology detection

Abstract

The present invention relates to the field of metrology calibration, and specifically discloses an intelligent calibration method, device and system for metrology detection, wherein the method includes matrix construction, graph conversion, multi-head self-attention processing, feature splicing and data calibration. The present invention creatively constructs a time-related adjacency matrix based on time series metrology data, learns the dynamic spatial dependency between nodes with a sparse related adjacency matrix, and constructs an undirected graph of historical spatiotemporal data with the time series metrology data for the extraction of spatiotemporal features, overcomes the over-smoothing problem of ordinary graph convolutional networks, introduces a multi-head self-attention mechanism to capture the diversified and specialized spatiotemporal dependencies within the time series metrology data, performs fine feature extraction on complex data, and realizes accurate prediction and data calibration. The intelligent calibration device and system for metrology detection are used to realize an intelligent calibration method for metrology detection.

Description

Technical Field

[0001] The present invention relates to the field of metrology calibration, and in particular to an intelligent calibration method, device and system for metrology detection. Background Art

[0002] Metrology and testing data calibration refers to the process of comparing and adjusting the data of measuring equipment, instruments or sensors to ensure that their output results are consistent with known standard values ​​or reference values. The purpose of calibration is to ensure the accuracy, reliability and traceability of measurement data, thereby ensuring that the measurement results meet specific quality requirements. Existing metrology and testing data are usually manually operated, with high calibration errors and low calibration efficiency. Existing intelligent calibration methods are difficult to consider the multidimensional characteristics of metrology data, and the calibration accuracy is low. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent calibration method, device and system for metrology detection. In view of the problem that manual operation has high calibration errors and low calibration efficiency, the present invention creatively adopts an intelligent calibration method for metrology detection, and calibrates the real data according to the predicted data by an intelligent prediction method, thereby improving the calibration efficiency and reducing the calibration error; in view of the problem that the existing intelligent calibration method is difficult to consider the multidimensional characteristics of the metrology data and has low calibration accuracy, the present invention creatively constructs a time-related adjacency matrix based on time series metrology data, uses a sparse related adjacency matrix to learn the dynamic spatial dependencies between nodes, and constructs an undirected graph of historical spatiotemporal data based on the time series metrology data for the extraction of spatiotemporal features, overcomes the over-smoothing problem of ordinary graph convolutional networks, introduces a multi-head self-attention mechanism to capture the diversified and specialized spatiotemporal dependencies within the time series metrology data, performs fine feature extraction on complex data, and realizes accurate prediction and data calibration.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent calibration method for measurement detection, the method comprising the following steps:

[0005] Step S1: Time correlation matrix construction: The time correlation matrix is ​​calculated based on the time series measurement data collected by all measurement sensors. The formula used is as follows:

[0006] ;

[0007] In the formula, and It is the time series measurement data collected by different measurement sensors. ∈ ,in is the number of all metering sensors, is the time step of the measurement sensor acquisition, and Represents the average value of time series measurement data, Represents the dot product operation between two data. represents the L2 norm operation of the data, represents the time correlation matrix;

[0008] Step S2: Time-dependent adjacency matrix construction: Use the time-dependent matrix to construct the time-dependent adjacency matrix. The formula used is as follows:

[0009] ;

[0010] In the formula, represents the standard deviation of the time correlation matrix, A preset hyperparameter threshold for adjusting the sparsity of the time-dependent adjacency matrix, represents the natural exponential function, represents the time-dependent adjacency matrix;

[0011] Step S3: Graph signal conversion, using the time series measurement data collected by all measurement sensors to construct an undirected graph of historical spatiotemporal data:

[0012] ;

[0013] ;

[0014] In the formula, Representative Sensors at time The collected measurement data, Represents at time The characteristic vector of represents the transpose of a vector, is the preset sampling interval threshold, that is Represents the feature vector In time Historical data, Represents an undirected graph of historical spatiotemporal data;

[0015] Step S4: multi-head self-attention processing of the time-dependent adjacency matrix, specifically including the following steps:

[0016] Step S41: Multi-head attention processing:

[0017] ;

[0018] In the formula, , and Represent the weight matrices for query, key, and value feature processing, respectively, , and Represent the time-dependent adjacency matrix query, key, and value features;

[0019] Step S42: Self-attention parameter matrix calculation:

[0020] ;

[0021] In the formula, represents the activation function, represents the modulus of the query feature, represents the self-attention parameter matrix;

[0022] Step S43: self-attention processing, using the self-attention parameter matrix and value features for self-attention processing:

[0023] ;

[0024] In the formula, represents the calculated self-attention feature;

[0025] Step S44: Multi-layer feedforward network processing and nonlinear activation:

[0026] ;

[0027] In the formula, represents the activation function, , and Represent the three-layer weight matrix of the feedforward network, Adjacent embedding features representing time series measurement data;

[0028] Step S5: Multi-head self-attention processing of historical spatiotemporal data undirected graph, repeating steps S41 to S44, with historical spatiotemporal data undirected graph As input, and sharing all calculation parameters, the spatiotemporal embedding features of time series measurement data are obtained;

[0029] Step S6: Feature concatenation and activation: concatenate the adjacent embedding features with the spatiotemporal embedding features and use the LSTM model for prediction output to obtain the time The corresponding predicted measurement data for the next sampling moment;

[0030] Step S7: Data calibration, comparing the predicted measurement data with the time series measurement data The actual data at the next sampling moment is compared and a calibration threshold is set. When the difference between the predicted metering data and the actual data exceeds the set calibration threshold, the metering sensor is adjusted using the predicted metering data as a reference value.

[0031] The present invention provides an intelligent calibration device for metrology detection, comprising a metrology sensor, a processor, and a memory, wherein the memory is used to store an application program for implementing an intelligent calibration method for metrology detection, the processor executes the application program and outputs the execution result, and the metrology sensor is used for collecting time series metrology data.

[0032] The present invention provides an intelligent calibration system for metrology detection, comprising a metrology sensor module, a data collection module, an intelligent calibration module, a data storage module and a user control module;

[0033] The metering sensor module captures the time series metering data with a metering sensor and sends the time series metering data to the data collection module;

[0034] The data collection module uses a metering sensor to obtain time series metering data and sends the time series metering data to the intelligent calibration module;

[0035] The intelligent calibration module extracts features and predicts measurement data for time series measurement data, and uses the predicted measurement data as a reference value to adjust the measurement sensor;

[0036] The data storage module stores the data of all modules;

[0037] The user control module allows the user to control and set parameters of all modules.

[0038] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0039] (1) In order to solve the problem of high calibration error and low calibration efficiency caused by manual operation, the present invention creatively adopts an intelligent calibration method for metrology detection, which uses an intelligent prediction method to calibrate the real data according to the predicted data, thereby improving the calibration efficiency and reducing the calibration error;

[0040] (2) In view of the problem that existing intelligent calibration methods have difficulty in considering the multi-dimensional characteristics of measurement data and have low calibration accuracy, the present invention creatively constructs a time-related adjacency matrix based on time series measurement data, uses a sparse related adjacency matrix to learn the dynamic spatial dependencies between nodes, and constructs an undirected graph of historical spatiotemporal data based on time series measurement data for the extraction of spatiotemporal features, overcoming the over-smoothing problem of ordinary graph convolutional networks, and introduces a multi-head self-attention mechanism to capture the diversified and specialized spatiotemporal dependencies within time series measurement data, perform fine feature extraction on complex data, and achieve accurate prediction and data calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1A schematic diagram of a flow chart of an intelligent calibration method for metrological detection provided by the present invention;

[0042] Figure 2 A schematic diagram of an intelligent calibration device for measurement and detection provided by the present invention;

[0043] Figure 3 A schematic diagram of an intelligent calibration system for metrology detection provided by the present invention.

[0044] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0046] Example 1, see Figure 1 The present invention provides an intelligent calibration method for measurement detection, which comprises the following steps:

[0047] Step S1: Time correlation matrix construction: The time correlation matrix is ​​calculated based on the time series measurement data collected by all measurement sensors. The formula used is as follows:

[0048] ;

[0049] In the formula, and It is the time series measurement data collected by different measurement sensors. ∈ ,in is the number of all metering sensors, is the time step of the measurement sensor acquisition, and Represents the average value of time series measurement data, Represents the dot product operation between two data. represents the L2 norm operation of the data, represents the time correlation matrix;

[0050] Step S2: Time-dependent adjacency matrix construction: Use the time-dependent matrix to construct the time-dependent adjacency matrix. The formula used is as follows:

[0051] ;

[0052] In the formula, represents the standard deviation of the time correlation matrix, A preset hyperparameter threshold for adjusting the sparsity of the time-dependent adjacency matrix, represents the natural exponential function, represents the time-dependent adjacency matrix;

[0053] Step S3: Graph signal conversion, using the time series measurement data collected by all measurement sensors to construct an undirected graph of historical spatiotemporal data:

[0054] ;

[0055] ;

[0056] In the formula, Representative Sensors at time The collected measurement data, Represents at time The characteristic vector of represents the transpose of a vector, is the preset sampling interval threshold, that is Represents the feature vector In time Historical data, Represents an undirected graph of historical spatiotemporal data;

[0057] Step S4: multi-head self-attention processing of the time-dependent adjacency matrix, specifically including the following steps:

[0058] Step S41: Multi-head attention processing:

[0059] ;

[0060] In the formula, , and Represent the weight matrices for query, key, and value feature processing, respectively, , and Represent the time-dependent adjacency matrix query, key, and value features;

[0061] Step S42: Self-attention parameter matrix calculation:

[0062] ;

[0063] In the formula, represents the activation function, represents the modulus of the query feature, represents the self-attention parameter matrix;

[0064] Step S43: self-attention processing, using the self-attention parameter matrix and value features for self-attention processing:

[0065] ;

[0066] In the formula, represents the calculated self-attention feature;

[0067] Step S44: Multi-layer feedforward network processing and nonlinear activation:

[0068] ;

[0069] In the formula, represents the activation function, , and Represent the three-layer weight matrix of the feedforward network, Adjacent embedding features representing time series measurement data;

[0070] Step S5: Multi-head self-attention processing of historical spatiotemporal data undirected graph, repeating steps S41 to S44, with historical spatiotemporal data undirected graph As input, and sharing all calculation parameters, the spatiotemporal embedding features of time series measurement data are obtained;

[0071] Step S6: Feature concatenation and activation: concatenate the adjacent embedding features with the spatiotemporal embedding features and use the LSTM model for prediction output to obtain the time The corresponding predicted measurement data for the next sampling moment;

[0072] Step S7: Data calibration, comparing the predicted measurement data with the time series measurement data The actual data at the next sampling moment is compared and a calibration threshold is set. When the difference between the predicted metering data and the actual data exceeds the set calibration threshold, the metering sensor is adjusted using the predicted metering data as a reference value.

[0073] By performing the above operations, in view of the problem that the existing intelligent calibration methods are difficult to consider the multi-dimensional characteristics of the measurement data and have low calibration accuracy, the present invention creatively constructs a time-related adjacency matrix based on the time series measurement data, and uses a sparse related adjacency matrix to learn the dynamic spatial dependencies between nodes, and constructs an undirected graph of historical spatiotemporal data with the time series measurement data for the extraction of spatiotemporal features, overcoming the over-smoothing problem of ordinary graph convolutional networks, introducing a multi-head self-attention mechanism to capture the diversified and specialized spatiotemporal dependencies within the time series measurement data, performing fine feature extraction on complex data, and achieving accurate prediction and data calibration.

[0074] Example 2, see Figure 2 The present invention provides an intelligent calibration device for metrology detection, including a metrology sensor, a processor, and a memory, wherein the memory is used to store an application program for implementing an intelligent calibration method for metrology detection, the processor executes the application program and outputs the execution result, and the metrology sensor is used for collecting time series metrology data.

[0075] Example 3, see Figure 3 , the present invention provides an intelligent calibration system for metrology detection, including a metrology sensor module, a data collection module, an intelligent calibration module, a data storage module and a user control module;

[0076] The metering sensor module captures the time series metering data with a metering sensor and sends the time series metering data to the data collection module;

[0077] The data collection module uses a metering sensor to obtain time series metering data and sends the time series metering data to the intelligent calibration module;

[0078] The intelligent calibration module extracts features and predicts measurement data for time series measurement data, and uses the predicted measurement data as a reference value to adjust the measurement sensor;

[0079] The data storage module stores the data of all modules;

[0080] The user control module allows the user to control and set parameters of all modules.

[0081] By performing the above operations, in order to solve the problem of high calibration error and low calibration efficiency caused by manual operation, the present invention creatively adopts an intelligent calibration method for measurement and detection, which uses an intelligent prediction method to calibrate the real data according to the predicted data, thereby improving the calibration efficiency and reducing the calibration error.

[0082] Embodiment 4: This embodiment is based on the above embodiment, and multi-head attention processing specifically includes the following steps:

[0083] Input: query (Q), key (K), and value (V);

[0084] Linear transformation generation head: The input query, key and value are mapped to multiple subspaces respectively through the trainable linear transformation matrix of the weight matrix;

[0085] Multi-head parallel attention calculation:

[0086] For each head, an attention score is calculated separately, which is calculated by the dot product of the query and the key.

[0087] Embodiment 5: This embodiment is based on the above embodiment. The metering sensor adopts an infrared distance measuring sensor, and the parameters are as follows:

[0088] Measuring range: 0.05 m (50 mm) to 20 m (20,000 mm);

[0089] Measuring accuracy: ±2 mm;

[0090] Working principle: Based on the time difference measurement principle;

[0091] Measuring speed: 100 times per second;

[0092] Output interface: RS232 digital interface;

[0093] Light source type: infrared laser;

[0094] Operating temperature range: -10°C to 50°C;

[0095] Protection level: IP65 protection level;

[0096] Power requirement: 24V DC;

[0097] Dimensions and weight: 100mm x 60mm x 40mm, weight approximately 200g.

[0098] Embodiment 6: This embodiment is based on the above embodiment. In step S5, the calculation parameters include:

[0099] Weight matrices for query, key, and value feature processing;

[0100] Activation function;

[0101] The three-layer weight matrix of the feedforward network;

[0102] Activation function.

[0103] Embodiment 7: This embodiment is based on the above embodiment. In step S6, vector concatenation specifically includes the following operations:

[0104] Prepare the two features to be concatenated, making sure they have the same dimensions;

[0105] To perform splicing operations:

[0106] Select row-by-row stitching mode;

[0107] Concatenate the two feature vectors together in order, and select the NumPy related function in Python to implement the concatenation operation.

[0108] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0109] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent calibration method for metrological detection, characterized in that: The method comprises the following steps: Step S1: Time correlation matrix construction: The time correlation matrix is ​​calculated based on the time series measurement data collected by all measurement sensors. The formula used is as follows: ； In the formula, and It is the time series measurement data collected by different measurement sensors. ∈ ,in is the number of all metering sensors, is the time step of the measurement sensor acquisition, and Represents the average value of time series measurement data, Represents the dot product operation between two data. represents the L2 norm operation of the data, represents the time correlation matrix; Step S2: Time-dependent adjacency matrix construction: Use the time-dependent matrix to construct the time-dependent adjacency matrix. The formula used is as follows: ； In the formula, represents the standard deviation of the time correlation matrix, A preset hyperparameter threshold for adjusting the sparsity of the time-dependent adjacency matrix, represents the natural exponential function, represents the time-dependent adjacency matrix; Step S3: Graph signal conversion, using the time series measurement data collected by all measurement sensors to construct an undirected graph of historical spatiotemporal data; Step S4: Multi-head self-attention processing of the time-related adjacency matrix to obtain the adjacency embedding features of the time series measurement data, specifically including the following steps: Step S41: Multi-head attention processing: ； In the formula, , and Represent the weight matrices for query, key, and value feature processing, respectively, , and Represent the time-dependent adjacency matrix query, key, and value features; Step S42: Self-attention parameter matrix calculation: ； In the formula, represents the activation function, represents the modulus of the query feature, represents the self-attention parameter matrix; Step S43: self-attention processing, using the self-attention parameter matrix and value features for self-attention processing: ； In the formula, represents the calculated self-attention feature; Step S44: Multi-layer feedforward network processing and nonlinear activation: ； In the formula, represents the activation function, , and Represent the three-layer weight matrix of the feedforward network, Adjacent embedding features representing time series measurement data; Step S5: Multi-head self-attention processing of historical spatiotemporal data undirected graph, repeat step S4, and use the historical spatiotemporal data undirected graph As input, and sharing all calculation parameters, the spatiotemporal embedding features of time series measurement data are obtained; Step S6: Feature concatenation and activation: concatenate the adjacent embedding features with the spatiotemporal embedding features and use the LSTM model for prediction output to obtain the time The corresponding predicted measurement data for the next sampling moment; Step S7: Data calibration, using the predicted metering data as a reference value to adjust the metering sensor.

2. An intelligent calibration device for metrology detection, used to implement the intelligent calibration method for metrology detection described in claim 1, characterized in that: It includes a metrological sensor, a processor, and a memory, wherein the memory is used to store an application program for implementing an intelligent calibration method for metrological detection, the processor executes the application program and outputs the execution result, and the metrological sensor is used for collecting time series metrological data.

3. An intelligent calibration system for metrology detection, used to implement the intelligent calibration method for metrology detection described in claim 1, characterized in that: It includes a metering sensor module, a data collection module, an intelligent calibration module, a data storage module and a central control module; The metering sensor module captures the time series metering data with a metering sensor and sends the time series metering data to the data collection module; The data collection module uses a metering sensor to obtain time series metering data and sends the time series metering data to the intelligent calibration module; The intelligent calibration module extracts features and predicts measurement data for time series measurement data, and uses the predicted measurement data as a reference value to adjust the measurement sensor; The data storage module stores the data of all modules; The central control module allows users to control and set parameters of all modules.

Images