A method for processing pirani data based on singular value decomposition

Abstract

This invention discloses a Pirani data processing method based on singular value decomposition (SVD), belonging to the field of Pirani data processing technology. The method includes placing a Pirani vacuum gauge in an experimental environment for data acquisition and processing. The acquired data is integrated into an m×n dimensional matrix A. Singular value decomposition is performed, and principal components are extracted. Principal components are selected, and matrix truncation and reconstruction are performed to achieve dimensionality reduction reconstruction with a fixed k value. Post-processing is then performed to obtain calibration curves, and pressure interpolation calculations are performed. Multi-temperature point calibration and storage of calibration tables are also included. This invention utilizes a Pirani data processing method based on singular value decomposition. The largest singular value and its corresponding basic waveform, and the second largest singular value and its corresponding waveform, capture the most important correction pattern. The third and subsequent singular values ​​typically correspond to data fluctuations caused by random interference, measurement noise, etc. By forcibly setting them to zero, the algorithm logic becomes simple and stable.

Description

Technical Field

[0001] This invention relates to the field of Pirani data processing technology, and in particular to a Pirani data processing method based on singular value decomposition. Background Technology

[0002] The Pirani vacuum gauge is a widely used measuring instrument in the field of low and medium vacuum. Its core working principle relies on two integrated sensors: a temperature sensor (channel 0, ch0) and a pressure sensor (channel 1, ch1). These two channels exhibit specific response characteristics. When the system pressure is kept constant, the reading of ch0 increases monotonically with the increase of ambient temperature, while the value of ch1 remains basically unchanged. Conversely, when the temperature conditions are stable, the reading of ch1 increases with the increase of system pressure, while the value of ch0 is unaffected.

[0003] To ensure the accuracy of factory measurements, each Pirani vacuum gauge undergoes precise pressure calibration. This calibration process requires uniformly changing the pressure within a certain range at multiple preset temperature points, while simultaneously recording the response value of ch1. Thus, at each specific temperature, a calibration curve characterizing the change of ch1 with pressure can be obtained, called the ch1 curve. Finally, multiple ch1 curves corresponding to different temperature points and their reference ch0 values ​​are integrated into a calibration data table and stored in the vacuum gauge's internal memory. During actual measurement, the device queries this calibration table using the real-time acquired ch0 and ch1 values ​​and calculates the current pressure value through interpolation.

[0004] Currently, the core challenge in this calibration process lies in the fact that, influenced by various random interference factors, even under strictly controlled temperature conditions, the ch1 curves obtained from multiple samplings are prone to numerical fluctuations and deviations at individual pressure points. To improve the accuracy of the final data, the current industry practice is to repeatedly sample the ch1 curve at the same temperature and then perform an arithmetic average of the data. This increases the amount of data acquisition and calculation, reducing the efficiency of Pirani vacuum gauge pressure calibration. Summary of the Invention

[0005] The purpose of this invention is to provide a Pirani data processing method based on singular value decomposition to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a Pirani data processing method based on singular value decomposition, comprising the following specific steps:

[0007] Step 1: Place the Pirani vacuum gauge in a temperature-controlled experimental environment for data acquisition and processing;

[0008] Step 2: Integrate the collected data into an m×n dimensional matrix A;

[0009] Step 3: Perform singular value decomposition and principal component extraction;

[0010] Step 4: Select principal components, perform matrix truncation and reconstruction, and achieve dimensionality reduction reconstruction with a fixed k value;

[0011] Step 5: Perform data post-processing to obtain the calibration curve and perform pressure interpolation calculations;

[0012] Step 6: Multi-temperature point calibration and storage of calibration tables.

[0013] Preferably, the data acquisition and data processing in step one includes the following steps:

[0014] S1.1: Select the Pirani vacuum gauge to be calibrated, place the Pirani vacuum gauge in a temperature-controlled experimental environment, stabilize the ambient temperature at the preset calibration temperature point, connect the vacuum gauge to the pressure control system, and adjust the pressure of the Pirani vacuum gauge within the operating range of the Pirani vacuum gauge.

[0015] S1.2: Under a fixed temperature, the pressure is increased stepwise, and the response values ​​of the pressure sensor ch1 at n preset pressure points are collected simultaneously to complete one calibration experiment;

[0016] S1.3: Repeat the above S1.2 operation steps m times to obtain m sets of original data sequences of ch1 as a function of pressure;

[0017] S1.4: Remove obvious outliers to ensure continuity and consistency in each data sequence. Align the m data sets according to the pressure point order so that the same column of data corresponds to the same pressure value.

[0018] Preferably, in step two, the collected m sets of data are integrated into an m×n dimensional matrix A. The m rows represent m independent calibration experiments, with each row corresponding to the complete ch1 response sequence in one experiment. The n columns represent n preset pressure sampling points, with each column corresponding to the ch1 measurement value at the same pressure point in different experiments. Matrix elements... Indicates the first In this experiment, the pressure was the first... The reading of ch1 at each sampling point.

[0019] Preferably, the formula for singular value decomposition in step three is:

[0020]

[0021] in, It is an m×m orthogonal matrix whose column vectors are left singular vectors. It is an m×n rectangular diagonal matrix, and its diagonal elements are... , , ... are singular values, and satisfy... ≥ ≥ ...≥0, It is an n×n orthogonal matrix whose column vectors are right singular vectors, representing the various potential fundamental waveforms that constitute the ch1 curve. This represents the matrix transpose operation, which swaps the rows and columns of a matrix.

[0022] Preferably, in step four, the principal component selection is based on the physical mechanism analysis and experimental verification of the Pirani vacuum gauge, fixing the selection of the first two largest singular values, with a selection quantity k=2, and the first singular value... The corresponding principal components represent the core common shape of the ch1 curve in all calibration experiments, and the second singular value. The corresponding principal components captured the main differences and correction patterns between experiments.

[0023] Preferably, the matrix truncation and reconstruction in step four includes the following steps:

[0024] S4.1: Construct the truncated diagonal matrix :reserve The first two singular values and The remaining singular values ​​are set to zero;

[0025] S4.2: Extracting the matrix :Pick The first two columns correspond to the first two left singular vectors;

[0026] S4.3: Extracting the matrix :Pick The first two columns correspond to the first two right singular vectors;

[0027] S4.4: Through , and The formula for the reconstructed filtered matrix is: .

[0028] Preferably, the calibration curve obtained in step five through data post-processing is derived from the reconstructed matrix. Extract each row of data to obtain m denoised ch1 calibration curves. Perform an arithmetic mean on the m curves to obtain the final ch1 calibration curve at that temperature.

[0029] Preferably, in step five, the pressure interpolation calculation stores the final ch1 curve at that temperature and the corresponding temperature sensor ch0 reference value in the calibration data table. During actual measurement, the calibration data table is queried and interpolation calculation is performed using the real-time collected ch0 and ch1 values ​​to obtain the current system pressure value.

[0030] Preferably, the multi-temperature point calibration in step six is ​​to repeat steps 1 to 5 above to complete the calibration of all preset temperature points, build a complete multi-temperature calibration database, and store the calibration database in the internal memory of the Pirani vacuum gauge to realize plug-and-play functionality and real-time high-precision pressure measurement of the device.

[0031] Preferably, constructing a complete multi-temperature calibration database includes the following steps:

[0032] S6.1: A three-level index structure of temperature, pressure and signal values ​​is adopted. Temperature values ​​are the first-level index, pressure point sequence is the second-level index, and signal values ​​are the third-level index. The storage units are the ch1 signal value array, the reference ch0 value and the data check code.

[0033] S6.2: Add a CRC32 checksum to the curve data of each temperature node and store it at the end of the data block. The checksum is automatically performed when reading. If data corruption is found, the backup copy reading process is triggered. Dual copies are stored for critical temperature nodes and distributed in different physical blocks of Flash.

[0034] S6.3: Store complete version information in the header of the calibration table, retain the parsing interface of the old version calibration table, support backward compatibility reading, and record version update logs to store the reasons for version iteration;

[0035] S6.4: In view of the characteristics of Flash storage, a wear leveling algorithm is adopted to distribute the calibration table data into different physical blocks and perform lossless compression on the optimized curve. The curve smoothness is used to store the difference sequence of adjacent pressure points and to establish SRAM cache area, Flash main storage and spare ROM area for partitioned storage.

[0036] S6.5: Establish a temperature index table, record the storage offset address of each temperature node in Flash, pre-calculate the difference coefficient matrix of adjacent temperature node curves, and store it in the calibration table.

[0037] The technical effects and advantages of this invention are as follows:

[0038] This invention utilizes a Pirani data processing method based on singular value decomposition to find the largest singular value. Its corresponding fundamental waveform represents the core shape of the ch1 curve, which is common and dominant in all calibration experiments, and the second largest singular value. The corresponding waveforms capture the most important correction modes, which are used to fine-tune the core shape to adapt to all experimental data. The third and subsequent singular values ​​usually correspond to data fluctuations caused by random interference, measurement noise, etc. By forcibly setting them to zero, these irrelevant and unstable components are actively removed from the reconstructed data, thereby achieving efficient noise filtering. The fixed k=2 criterion avoids complex threshold calculations or energy percentage judgments, making the algorithm logic simple, stable, and easy to implement and deploy in embedded systems. Attached Figure Description

[0039] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention, but do not constitute a limitation thereof. In the drawings:

[0040] Figure 1 This is a flowchart of the Pirani data processing method based on singular value decomposition according to the present invention;

[0041] Figure 2 This is a flowchart of the reconstructed calibration curve of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] This invention provides, for example Figures 1-2 The Pirani data processing method based on singular value decomposition, as shown, includes the following specific steps:

[0044] Step 1: Place the Pirani vacuum gauge in a temperature-controlled experimental environment for data acquisition and processing;

[0045] Step 2: Integrate the collected data into an m×n dimensional matrix A;

[0046] Step 3: Perform singular value decomposition and principal component extraction;

[0047] Step 4: Select principal components, perform matrix truncation and reconstruction, and achieve dimensionality reduction reconstruction with a fixed k value;

[0048] Step 5: Perform data post-processing to obtain the calibration curve and perform pressure interpolation calculations;

[0049] Step 6: Multi-temperature point calibration and storage of calibration tables.

[0050] Furthermore, the data acquisition and processing in step one includes the following steps:

[0051] S1.1: Select the Pirani vacuum gauge to be calibrated and place it in a temperature-controlled experimental environment, such as a constant temperature chamber, to stabilize the ambient temperature at the preset calibration temperature points, such as 0℃, 20℃, and 40℃. Connect the vacuum gauge to the pressure control system, ensuring the pressure from the pressure control system is within the operating range of the Pirani vacuum gauge, for example, at 1×10⁻⁻⁶. 5 The range is from mbar to 1000 mbar;

[0052] S1.2: At a fixed temperature, control the pressure to increase in a stepwise manner, for example from 1×10⁻ 5 The pressure is increased at a constant rate from mbar to 1000 mbar, and the response values ​​of the pressure sensor ch1 at n preset pressure points are collected simultaneously to complete one calibration experiment.

[0053] S1.3: Repeat the above S1.2 operation steps m times, m≥3, to ensure the statistical significance of the data, and obtain m sets of original data sequences of ch1 changing with pressure;

[0054] S1.4: Remove obvious outliers, such as sudden changes in values ​​caused by momentary equipment failure, to ensure the continuity and consistency of each data sequence. Align the m data sets according to the pressure point order so that the data in the same column correspond to the same pressure value.

[0055] Furthermore, in step two, the collected m sets of data are integrated into an m×n dimensional matrix A. The m rows represent m independent calibration experiments, with each row corresponding to the complete ch1 response sequence in one experiment. The n columns represent n preset pressure sampling points, with each column corresponding to the ch1 measurement value at the same pressure point in different experiments. Matrix elements... Indicates the first In this experiment, the pressure was the first... The reading of ch1 at each sampling point.

[0056] Furthermore, the formula for singular value decomposition in step three is:

[0057]

[0058] in, It is an m×m orthogonal matrix whose column vectors are left singular vectors. It is an m×n rectangular diagonal matrix, and its diagonal elements are... , , ... are singular values. Singular values ​​are non-negative real numbers generated by a matrix in Singular Value Decomposition (SVD). They are one of the essential eigenvalues ​​of a matrix, used to quantify the energy or importance of different components within the matrix, and satisfy the following conditions: ≥ ≥ ...≥0, these values ​​quantify the energy or importance of the corresponding component in the original data. It is an n×n orthogonal matrix whose column vectors are right singular vectors, representing the various potential fundamental waveforms that constitute the ch1 curve. This represents the matrix transpose operation, which interchanges the rows and columns of a matrix. For example, if... If it is an n×n matrix, then That is to The The line becomes the first Column, number The column becomes the first The dimension remains n×n.

[0059] Specifically, in step four, the principal component selection is based on the physical mechanism analysis and experimental verification of the Pirani vacuum gauge. The two largest singular values ​​are fixed, with a selection quantity k=2. The first singular value... The corresponding principal components represent the core common shape of the ch1 curve in all calibration experiments, and the second singular value. The corresponding principal components captured the main differences and correction patterns between experiments, while the outliers ranked third and below were mainly caused by interference factors such as random noise and measurement errors.

[0060] In particular, the matrix truncation and reconstruction in step four includes the following steps:

[0061] S4.1: Construct the truncated diagonal matrix :reserve The first two singular values and The remaining singular values ​​are set to zero;

[0062] S4.2: Extracting the matrix :Pick The first two columns correspond to the first two left singular vectors;

[0063] S4.3: Extracting the matrix :Pick The first two columns correspond to the first two right singular vectors;

[0064] S4.4: Through , and The formula for the reconstructed filtered matrix is: .

[0065] Furthermore, the calibration curve obtained in step five through data post-processing is derived from the reconstructed matrix. Extract each row of data to obtain m denoised ch1 calibration curves. Perform an arithmetic mean on the m curves to obtain the final ch1 calibration curve at that temperature. In step five, the pressure interpolation calculation stores the final ch1 curve at that temperature and the corresponding temperature sensor ch0 reference value into the calibration data table. During actual measurement, the calibration data table is queried and interpolation calculation is performed by using the real-time collected ch0 and ch1 values ​​to obtain the pressure value of the current system.

[0066] In particular, step six, multi-temperature point calibration, involves repeating steps 1 to 5 above to complete the calibration of all preset temperature points, such as 0℃, 20℃, and 40℃, thus constructing a complete multi-temperature calibration database. This calibration database is then stored in the internal memory of the Pirani vacuum gauge, enabling plug-and-play functionality and real-time high-precision pressure measurement. Constructing the complete multi-temperature calibration database includes the following steps:

[0067] S6.1: A three-level index structure of temperature, pressure and signal values ​​is adopted. Temperature values ​​are the first-level index, pressure point sequence is the second-level index, and signal values ​​are the third-level index. The storage units are the ch1 signal value array, the reference ch0 value and the data check code.

[0068] S6.2: Add a CRC32 checksum to the curve data for each temperature node. The CRC32 checksum is an error detection algorithm widely used in data transmission and storage. It generates a 32-bit (4-byte) checksum by performing polynomial operations on the original data. This checksum is used to verify whether the data has been damaged, tampered with, or lost during transmission or storage. The CRC32 checksum is named for its fixed length of 32 bits. It is one of the most commonly used verification methods in embedded systems, network communication, and file storage. It is stored at the end of the data block and is automatically checked when read. If data corruption is found, the backup copy reading process is triggered. Dual copies are stored at critical temperature nodes and distributed in different physical blocks of Flash. Flash is a non-volatile semiconductor memory.

[0069] S6.3: Store complete version information in the header of the calibration table, retain the parsing interface of the old version calibration table, support backward compatibility reading, provide incremental update mode, only replace the newly added or optimized temperature node data, without rewriting the whole table, and record the version update log to store the reason for the version iteration.

[0070] S6.4: Targeting the characteristics of Flash storage, a wear leveling algorithm is adopted to distribute the calibration table data into different physical blocks and perform lossless compression on the optimized curve. Utilizing the smoothness of the curve, the difference sequence between adjacent pressure points is stored. SRAM cache, Flash main storage, and backup ROM are partitioned for storage. The SRAM cache stores frequently used data in the normal temperature range, such as data in the 10℃-30℃ temperature range, enabling microsecond-level reading. Flash main storage stores data across the entire temperature range, which is loaded into memory as needed. The backup ROM contains the default baseline curve, based on the average calibration model of 100 sensors, serving as a final fallback. Fallback is a fallback mechanism, a fault-tolerant strategy in system design. When the main function, main data, or main service fails, becomes unavailable, or is abnormal, the system automatically switches to the pre-prepared backup plan to ensure the continuity and stability of core functions and avoid complete system paralysis due to a single point of failure.

[0071] S6.5: Establish a temperature index table to record the storage offset address of each temperature node in Flash. Support binary search of adjacent temperature intervals based on real-time ch0 value. Pre-calculate the difference coefficient matrix of adjacent temperature node curves and store it in the calibration table. During actual measurement, directly call the pre-calculated coefficients for linear interpolation, reducing the real-time calculation load by about 40%.

[0072] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A Pirani data processing method based on singular value decomposition, characterized in that: The specific steps include the following: Step 1: Place the Pirani vacuum gauge in a temperature-controlled experimental environment for data acquisition and processing; Step 2: Integrate the collected data into an m×n dimensional matrix A; Step 3: Perform singular value decomposition and principal component extraction; Step 4: Select principal components, perform matrix truncation and reconstruction, and achieve dimensionality reduction reconstruction with a fixed k value; Step 5: Perform data post-processing to obtain the calibration curve and perform pressure interpolation calculations; Step 6: Multi-temperature point calibration and storage of calibration tables.

2. The Pirani data processing method based on singular value decomposition according to claim 1, characterized in that: The data acquisition and data processing in step one includes the following steps: S1.1: Select the Pirani vacuum gauge to be calibrated, place the Pirani vacuum gauge in a temperature-controlled experimental environment, stabilize the ambient temperature at the preset calibration temperature point, connect the vacuum gauge to the pressure control system, and adjust the pressure of the Pirani vacuum gauge within the operating range of the Pirani vacuum gauge. S1.2: Under a fixed temperature, the pressure is increased stepwise, and the response values ​​of the pressure sensor ch1 at n preset pressure points are collected simultaneously to complete one calibration experiment; S1.3: Repeat the above S1.2 operation steps m times to obtain m sets of original data sequences of ch1 as a function of pressure; S1.4: Remove obvious outliers to ensure continuity and consistency in each data sequence. Align the m data sets according to the pressure point order so that the same column of data corresponds to the same pressure value.

3. The Pirani data processing method based on singular value decomposition according to claim 1, characterized in that: In step two, the collected m sets of data are integrated into an m×n dimensional matrix A. The m rows represent m independent calibration experiments, with each row corresponding to the complete ch1 response sequence in one experiment. The n columns represent n preset pressure sampling points, with each column corresponding to the ch1 measurement value at the same pressure point in different experiments. Matrix elements... Indicates the first In this experiment, the pressure was the first... The reading of ch1 at each sampling point.

4. The Pirani data processing method based on singular value decomposition according to claim 3, characterized in that: The formula for singular value decomposition in step three is as follows: in, It is an m×m orthogonal matrix whose column vectors are left singular vectors. It is an m×n rectangular diagonal matrix, and its diagonal elements are... , , ... are singular values, and satisfy... ≥ ≥ ...≥0, It is an n×n orthogonal matrix whose column vectors are right singular vectors, representing the various potential fundamental waveforms that constitute the ch1 curve. This represents the matrix transpose operation, which swaps the rows and columns of a matrix.

5. The Pirani data processing method based on singular value decomposition according to claim 4, characterized in that: In step four, the principal component selection is based on the physical mechanism analysis and experimental verification of the Pirani vacuum gauge. The two largest singular values ​​are fixed and selected, with a selection quantity k=2. The first singular value... The corresponding principal components represent the core common shape of the ch1 curve in all calibration experiments, and the second singular value. The corresponding principal components captured the main differences and correction patterns between experiments.

6. The Pirani data processing method based on singular value decomposition according to claim 5, characterized in that: Step four, matrix truncation and reconstruction, includes the following steps: S4.1: Construct the truncated diagonal matrix :reserve The first two singular values and The remaining singular values ​​are set to zero; S4.2: Extracting the matrix :Pick The first two columns correspond to the first two left singular vectors; S4.3: Extracting the matrix :Pick The first two columns correspond to the first two right singular vectors; S4.4: Through , and The formula for the reconstructed filtered matrix is: .

7. The Pirani data processing method based on singular value decomposition according to claim 6, characterized in that: The calibration curve obtained in step five is derived from the reconstructed matrix. Extract each row of data to obtain m denoised ch1 calibration curves. Perform an arithmetic mean on the m curves to obtain the final ch1 calibration curve at that temperature.

8. The Pirani data processing method based on singular value decomposition according to claim 7, characterized in that: In step five, the pressure interpolation calculation stores the final ch1 curve at the specified temperature and the corresponding ch0 reference value of the temperature sensor in the calibration data table. During actual measurement, the calibration data table is queried and interpolation calculation is performed using the real-time collected ch0 and ch1 values ​​to obtain the current system pressure value.

9. The Pirani data processing method based on singular value decomposition according to claim 8, characterized in that: In step six, the multi-temperature point calibration is to repeat steps 1 to 5 above to complete the calibration of all preset temperature points, build a complete multi-temperature calibration database, and store the calibration database in the internal memory of the Pirani vacuum gauge to realize the device's plug-and-play functionality and real-time high-precision pressure measurement.

10. The Pirani data processing method based on singular value decomposition according to claim 9, characterized in that: The construction of a complete multi-temperature calibration database includes the following steps: S6.1: A three-level index structure of temperature, pressure and signal values ​​is adopted. Temperature values ​​are the first-level index, pressure point sequence is the second-level index, and signal values ​​are the third-level index. The storage units are the ch1 signal value array, the reference ch0 value and the data check code. S6.2: Add a CRC32 checksum to the curve data of each temperature node and store it at the end of the data block. The checksum is automatically performed when reading. If data corruption is found, the backup copy reading process is triggered. Dual copies are stored for critical temperature nodes and distributed in different physical blocks of Flash. S6.3: Store complete version information in the header of the calibration table, retain the parsing interface of the old version calibration table, support backward compatibility reading, and record version update logs to store the reasons for version iteration; S6.4: In view of the characteristics of Flash storage, a wear leveling algorithm is adopted to distribute the calibration table data into different physical blocks and perform lossless compression on the optimized curve. The curve smoothness is used to store the difference sequence of adjacent pressure points and to establish SRAM cache area, Flash main storage and spare ROM area for partitioned storage. S6.5: Establish a temperature index table, record the storage offset address of each temperature node in Flash, pre-calculate the difference coefficient matrix of adjacent temperature node curves, and store it in the calibration table.

Images