Semiconductor gas sensor drift fault diagnosis method based on information gain

Abstract

The invention provides a semiconductor gas sensor drift fault diagnosis method based on information gain, and relates to the technical field of semiconductor gas sensor drift fault diagnosis. Time domain temperature control response signals and drift fault signals in the temperature control heating process of the semiconductor gas sensor are collected, and a sample library is established; dividing a time domain temperature control response signal into a plurality of time windows by applying a sliding window method, and calculating information gains of the time windows; manually judging the operation state of the semiconductor gas sensor to obtain an operation state label of the semiconductor gas sensor; establishing a sample database by taking the information gain and the operation state tag as samples; and fitting the samples in the sample database into a function relationship between information gain and an operation state label through a support vector regression (SVM) algorithm, obtaining temperature control response data of the semiconductor gas sensor, performing information gain calculation, obtaining the operation state label of the semiconductor gas sensor, and completing fault diagnosis of the semiconductor gas sensor.

Description

Technical Field

[0001] The present invention relates to the technical field of semiconductor gas sensor drift fault diagnosis, and in particular to a semiconductor gas sensor drift fault diagnosis method based on information gain. Background Art

[0002] During the gas monitoring process in a complex gas environment, changes in environmental factors and gas properties often have a significant impact on the performance of semiconductor gas sensors. In particular, semiconductor gas sensors may exhibit drift faults, resulting in unstable reactions with gas molecules and thus causing drift phenomena in feedback signals.

[0003] This drift affects the information accuracy carried by the response signals of semiconductor gas sensors. Different gas types and sensor models have different signal response characteristics, making the monitoring process more complex. In the same complex gas environment, when the gas concentration changes, the response signals of semiconductor gas sensors usually show a fluctuating increase or decrease. However, when the sensor fails, the feedback signal will show a drift-like increase or decrease, presenting irregular fluctuations. When both gas concentration changes and sensor drift faults occur simultaneously, how to distinguish these two different signal fluctuations becomes a major challenge in the monitoring process. This makes it more difficult to accurately judge the operating state of semiconductor gas sensors and the validity of feedback signals in real-time monitoring. Theoretically, for the monitoring requirements in a complex gas environment, the fluctuations of feedback response signals should be monitored in real-time according to the characteristics of semiconductor gas sensors and the properties of the gases to be monitored. By judging these changes, it is possible to effectively evaluate whether the sensor is in a stable operating state, thereby improving the accuracy and reliability of gas monitoring. This view has been widely recognized in laboratory research and engineering applications. However, in actual complex gas environment monitoring, experienced operators usually rely on feedback signals under laboratory conditions to judge the current operating state of gas sensors.

[0004] Nevertheless, this method still has limitations. It is difficult to have a general judgment standard in the case of multiple differences in the response signals of different gas sensors. In addition, there is currently a lack of an effective artificial intelligence method that can assist operators in judging the actual operating state of current sensors according to real-time gas conditions. Especially when both gas concentration changes and sensor drift faults occur simultaneously, it is difficult for operators to make a quick judgment, resulting in impacts on the accuracy and timeliness of monitoring effects. Therefore, developing an artificial intelligence method that can self-decide and judge the operating state of sensors in real-time is of great research significance and application prospects for improving the monitoring ability of semiconductor gas sensors in complex environments and ensuring their stable operation. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a drift fault diagnosis method for semiconductor gas sensors based on information gain, which solves the problem that the existing drift fault diagnosis of semiconductor gas sensors mainly relies on manual expert judgment, especially in the case of concentration fluctuations, and provides a fault diagnosis algorithm for semiconductor gas sensors based on information gain. Information gain calculates the information gain between signals by designing a joint kernel function to quantify the distribution difference of signals in the non-linear feature characterization mode, breaking through the limitation of the insufficient dynamic adaptability of the linear model to complex systems. The present invention judges the operating state of the semiconductor gas sensor under the current gas environment conditions by calculating the information gain of the one-dimensional time-domain signal in combination with a machine learning algorithm.

[0006] A drift fault diagnosis method for semiconductor gas sensors based on information gain, comprising the following steps:

[0007] Step 1: Collect the time-domain temperature control response signal and the drift fault signal during the temperature control heating process of the semiconductor gas sensor;

[0008] Step 2: Establish a sample library according to the time-domain temperature control response signal and the drift fault signal collected in Step 1;

[0009] Step 3: Apply the sliding window method to the time-domain temperature control response signal, set the size and sliding step of the sliding window, and divide the time-domain signal into multiple time windows;

[0010] Step 3.1: Take the collected time-domain temperature control response signal as the current state data: S cur ={x1,x2,x3,…,x t}, where t is the current time, and x t is the value of the time-domain temperature control response signal at time t, and the size of the sliding window is set to N;

[0011] Step 3.2: Dynamic state reference update, using the sliding window method, the reference state is the nearest N data points, and N is the size of the sliding window:

[0012] S ref ={x t-N+1 ,x t-N+2 ,x t-N+3 ,…,x t}

[0013] Step 4: Calculate the information gain for each time window;

[0014] Step 4.1: Combine the local sensitivity of the Gaussian kernel K rbf and the global characteristics of the polynomial kernel K poly to perform kernel function fusion, specifically as follows:

[0015] The Gaussian kernel K rbf function is as follows:

[0016]

[0017] The polynomial kernel K poly is defined as follows:

[0018] K poly (x, y) = (x T y + c) d

[0019] Kernel function fusion:

[0020] K combined (x, y) = α·K rbf (x, y) + β·K poly (x, y)

[0021] where α and β are weight coefficients, and their optimal values are determined by grid search; x is the input of the current state data, y is the input of the reference state data, c is the polynomial offset, and σ is the Gaussian bandwidth;

[0022] Step 4.2: Calculate the probability density of the reference state, and use the kernel function to calculate the probability densities of the reference state and the current state;

[0023] The probability density of the reference state is as follows:

[0024]

[0025] The probability density of the current state is as follows:

[0026]

[0027] where: and are the data points of the reference state and the current state respectively, N ref is the total number of data points collected for the reference state, i is the i-th data, and N current is the total number of data points collected for the current state.

[0028] Step 4.3: Calculate the non-linear information entropy, including the reference state information entropy and the current state information entropy:

[0029] The reference state information entropy is as follows:

[0030]

[0031] The current state information entropy is as follows:

[0032]

[0033] where N refThe total number of data points collected for the reference state, i is the i-th data, N current The total number of data points collected for the current state.

[0034] Step 4.4: Calculate the information gain, specifically as follows:

[0035] SIG = H kernel (S ref ) - H kernel (S current )

[0036] Step 5: Repeat Steps 1 - 4 for the temperature control response signals and drift fault signals of all gases with different concentrations and types collected in sequence;

[0037] Step 6: Manually judge the operating state of the semiconductor gas sensor to obtain the actual operating state label of the semiconductor gas sensor;

[0038] The operating state label includes: the normal state label is 00, the fault operating state label is 01, and the fault and concentration change operating label is 11;

[0039] Step 7: Establish a sample database according to the information gain and the operating state label of the semiconductor gas sensor as samples;

[0040] Step 8: Fit the function relationship between the information gain and the operating state label by the support vector regression algorithm SVM for the samples in the sample database and save it in the form of a function;

[0041] Step 8.1: Shuffle the order of all samples, randomly sort them, take some samples as training data, and the remaining samples as test data;

[0042] Step 8.2: Fit the training data by the support vector regression algorithm SVM and test the fitting result. When the test error is less than 0.1%, retain the data, and the obtained function relationship is as follows:

[0043] state = SVM(SIG)

[0044] where: state is the operating state of the semiconductor gas sensor.

[0045] Step 9: Obtain the temperature control response data of the semiconductor gas sensor and calculate the information gain according to Step 4.

[0046] Step 10: Use the information gain obtained in Step 9 as the input, and calculate the operating state label of the semiconductor gas sensor through the function obtained in Step 8 to complete the fault diagnosis of the semiconductor gas sensor.

[0047] The beneficial effects of adopting the above technical solutions are as follows:

[0048] The present invention provides a drift fault diagnosis method for semiconductor gas sensors based on information gain. According to the combination of historical data and theoretical analysis, the characteristic differences of non-linear dynamic information gain can more accurately reflect the operating state of the response of semiconductor sensors. At the same time, an SVM learning machine is used to model the fault identification of the operating state, establish the relationship between non-linear dynamic information gain and the operating fault state of semiconductor gas sensors, realize the judgment of the operating state of semiconductor gas sensors, change the current situation of judging the operating state of semiconductor gas sensors by manual experience calibration, and improve the work efficiency of gas detection personnel. At the same time, reduce manual intervention and avoid unnecessary errors.

[0049] This method is driven by the response data during the temperature control heating process of semiconductor gas sensors, combines non-linear spatial gain calculation and artificial intelligence algorithms, and can accurately analyze and judge the real-time operating state of semiconductor gas sensors. Through this method, operators can more accurately evaluate the operating conditions of the gas system, detect sensor faults in a timely manner, and provide strong support for fault diagnosis and maintenance decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is the overall flowchart of the drift fault diagnosis method for semiconductor gas sensors of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The following combines the drawings and embodiments to further describe in detail the specific embodiments of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0052] A drift fault diagnosis method for semiconductor gas sensors based on information gain is described in detail below in combination with Figure 1 , and the specific implementation of the present invention is described in detail, including the following steps:

[0053] Step 1: Collect the time-domain temperature control response signal and drift fault signal during the temperature control heating process of the semiconductor gas sensor;

[0054] Step 2: Establish a sample library according to the time-domain temperature control response signal and drift fault signal collected in Step 1;

[0055] Step 3: Apply the sliding window method to the time-domain temperature control response signal, set the size and sliding step of the sliding window, and divide the time-domain signal into multiple time windows;

[0056] Step 3.1: Take the collected time-domain temperature control response signal as the current state data: S cur ={x1, x2, x3,..., x t}, where t is the current time, and x t is the time-domain temperature control response signal value at time t, and the size of the sliding window is set to N; in this embodiment, N is 10;

[0057] Step 3.2: Dynamic state reference update. Using the sliding window method, the reference state is the last N data points, where N is the size of the sliding window:

[0058] S ref ={x t-N+1 , x t-N+2 , x t-N+3 , …, x t}

[0059] Step 4: For each time window, calculate its information gain;

[0060] Step 4.1: Combine the local sensitivity of the Gaussian kernel K rbf and the global characteristics of the polynomial kernel K poly to perform kernel function fusion, as follows:

[0061] The Gaussian kernel K rbf function is as follows:

[0062]

[0063] The polynomial kernel K poly function is as follows:

[0064] K poly (x, y)=(x T y + c) d

[0065] Kernel function fusion:

[0066] K combined (x, y)=α·K rbf (x, y)+β·K poly (x, y)

[0067] where: α and β are weight coefficients, and their optimal values are determined by grid search, x is the current state data input, y is the reference state data input, c is the polynomial offset, and σ is the Gaussian bandwidth;

[0068] Step 4.2: Calculate the probability density of the reference state, and use the kernel function to calculate the probability density of the reference state and the current state;

[0069] The probability density of the reference state is as follows:

[0070]

[0071] The probability density of the current state is as follows:

[0072]

[0073] Wherein: and are the data points of the reference state and the current state respectively, N ref is the total number of data points collected in the reference state, i is the i-th data, N current is the total number of data points collected in the current state.

[0074] Step 4.3: Calculate the non-linear information entropy, including the reference state information entropy and the current state information entropy:

[0075] The reference state information entropy is as follows:

[0076]

[0077] The current state information entropy is as follows:

[0078]

[0079] Wherein: N ref is the total number of data points collected in the reference state, i is the i-th data, N current is the total number of data points collected in the current state.

[0080] Step 4.4: Calculate the information gain, specifically as follows:

[0081] SIG = H kernel (S ref ) - H kernel (S current )

[0082] Step 5: Repeat Steps 1 - 4 for the temperature control response signals and drift fault signals of all gases with different concentrations and different types collected in sequence;

[0083] Step 6: Manually judge the operating state of the semiconductor gas sensor to obtain the actual operating state label of the semiconductor gas sensor;

[0084] The operating state labels include: the normal state label is 00, the faulty operating state label is 01, and the faulty and concentration-changing operating label is 11;

[0085] Step 7: Establish a sample database based on the information gain and the operating state labels of the semiconductor gas sensor as samples;

[0086] Step 8: Fit the function relationship between the information gain and the operating state labels through the support vector regression algorithm SVM for the samples in the sample database and save it in the form of a function;

[0087] Step 8.1: shuffle the order of all samples, sort them randomly, take some samples as training data, and the remaining samples as test data; in this embodiment, take the first four fifths of the samples as training data, and the last one fifth as test data;

[0088] Step 8.2: Fit the training data using the support vector regression algorithm SVM and test the fitting results. When the test error is less than 0.1%, retain the data and obtain the following functional relationship:

[0089] state = SVM(SIG)

[0090] Wherein: state is the operating state of the semiconductor gas sensor.

[0091] Step 9: Obtain the temperature control response data of the semiconductor gas sensor and calculate the information gain according to step 4.

[0092] Step 10: Use the information gain obtained in step 9 as input, calculate the operating status label of the semiconductor gas sensor through the function obtained in step 8, and complete the fault diagnosis of the semiconductor gas sensor.

[0093] Embodiment 1:

[0094] Taking the actual situation of gas 1 and semiconductor gas sensor 1 as an example, the test is performed to obtain the information gain SIG calculation value of the signal information 1 to be tested and the operating status label of the semiconductor gas sensor 1.

[0095] The information gain SIG calculated value of gas 1, the operating status label of semiconductor gas sensor 1, the gain SIG calculated value of the signal information to be measured 1, and the state operation result of semiconductor gas sensor 1 are shown in Table 1, Table 2, Table 3 and Table 4 respectively:

[0096] Table 1 Information gain SIG calculated for gas 1

[0097]

[0098] Table 2 Operation status label of semiconductor gas sensor 1

[0099]

[0100] Table 3 Information gain SIG calculation value of the signal under test 1

[0101]

[0102] Table 4 Status operation results of the semiconductor gas sensor 1 in this embodiment

[0103]

[0104] Example 2:

[0105] Taking the sum of Gas 2 and the actual situation of the semiconductor gas sensor 2 as an example for testing, the calculated value of the information gain SIG of the signal to be measured 2 and the operation status label of the semiconductor gas sensor 2 are obtained.

[0106] The calculated value of the information gain SIG of Gas 2, the operation status label of the semiconductor gas sensor 2, the calculated value of the information gain SIG of the signal to be measured 2, and the operation result of the sensor status of Semiconductor Gas 2 are shown in Table 5, Table 6, Table 7, and Table 8 respectively:

[0107] Table 5 Calculated value of the information gain SIG of Gas 2

[0108]

[0109] Table 6 Operation status label of the semiconductor gas sensor 2

[0110]

[0111] Table 7 Calculated value of the information gain SIG of the signal to be measured 2

[0112]

[0113] Table 8 Operation result of the status of the semiconductor gas sensor 2

[0114]

[0115] The above description is only the preferred embodiments of the present disclosure and the description of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) disclosed in the embodiments of the present disclosure that have similar functions.

Claims

1. A semiconductor gas sensor drift fault diagnosis method based on information gain, characterized in that It includes the following steps: Step 1: Collect the time-domain temperature control response signal and drift fault signal during the temperature control heating process of the semiconductor gas sensor; Step 2: Establish a sample library based on the time-domain temperature control response signal and drift fault signal collected in Step 1; Step 3: Apply the sliding window method to the time-domain temperature control response signal, set the size and sliding step of the sliding window, and divide the time-domain signal into multiple time windows; Step 4: For each time window, calculate its information gain; Step 5: Repeat Steps 1-4 in sequence for the temperature control response signals and drift fault signals of all gases with different concentrations and types collected; Step 6: Manually judge the operating state of the semiconductor gas sensor to obtain the actual operating state label of the semiconductor gas sensor; Step 7: Establish a sample library database with the information gain and the operating state label of the semiconductor gas sensor as samples; Step 8: Fit the function relationship between the information gain and the operating state label for the samples in the sample library database through the support vector regression algorithm SVM and save it in the form of a function; Step 9: Obtain the temperature control response data of the semiconductor gas sensor and calculate the information gain according to Step 4; Step 10: Use the information gain obtained in Step 9 as the input, calculate the operating state label of the semiconductor gas sensor through the function obtained in Step 8, and complete the fault diagnosis of the semiconductor gas sensor.

2. The drift fault diagnosis method of a semiconductor gas sensor based on information gain according to claim 1, characterized in that The said Step 3 includes the following steps: Step 3.1: Take the collected time-domain temperature control response signal as the current state data: S cur ={x1, x2, x3, …, x t}, where t is the current time, and x t is the value of the time-domain temperature control response signal at time t. Set the size of the sliding window to N; Step 3.2: Dynamic state reference update, using the sliding window method, the reference state is the nearest N data points, and N is the size of the sliding window: S ref = {x t-N+1 , x t-N+2 , x t-N+3 , …, x t}。 3. A drift fault diagnosis method for semiconductor gas sensors based on information gain according to claim 1, characterized in that The said Step 4 includes the following steps: Step 4.1: Combine the local sensitivity of the Gaussian kernel K rbf and the global characteristics of the polynomial kernel K poly to perform kernel function fusion, as follows: The Gaussian kernel K rbf is defined as follows: The polynomial kernel K poly is as follows: K poly (x,y) = (x T y + c) d Kernel function fusion: K combined (x,y) = α·K rbf (x,y) + β·K poly (x,y) Where: α and β are weight coefficients, and the optimal values are determined through grid search, x is the current state data input, y is the reference state data input, c is the polynomial offset, and σ is the Gaussian bandwidth; Step 4.2: Calculate the probability density of the reference state, and use the kernel function to calculate the probability density of the reference state and the current state; The probability density of the said reference state is as follows: The probability density of the said current state is as follows: Wherein: and are data points of the reference state and the current state respectively, N ref is the total number of data points collected in the reference state, i is the i-th data, N current is the total number of data points collected in the current state; Step 4.3: Calculate the non-linear information entropy, including the reference state information entropy and the current state information entropy: The reference state information entropy is as follows: The current state information entropy is as follows: Where: N ref is the total number of data points collected in the reference state, i is the i-th data, and N current is the total number of data points collected in the current state; Step 4.4: Calculate the information gain.

4. A drift fault diagnosis method for a semiconductor gas sensor based on information gain according to claim 3, characterized in that, The information gain described in Step 4.4 is specifically as follows: SIG = H kernel (S ref ) - H kernel (S current )。 5. A drift fault diagnosis method for a semiconductor gas sensor based on information gain according to claim 1, characterized in that, The operating state labels described in Step 6 include: the normal state label is 00, the fault operating state label is 01, and the fault and concentration change operating label is 11.

6. A drift fault diagnosis method for a semiconductor gas sensor based on information gain according to claim 1, characterized in that The said Step 8 includes the following steps: Step 8.1: Shuffle the order of all samples, randomly sort them, take part of the samples as training data, and the remaining samples as test data; Step 8.2: Fit the training data through the support vector regression algorithm SVM and test the fitting result. When the test error is less than 0.1%, retain the data to obtain the function relationship.

7. A drift fault diagnosis method for a semiconductor gas sensor based on information gain according to claim 6, characterized in that, The said function relationship is as follows: state = SVM(SIG) Where: state is the operating state of the semiconductor gas sensor.

Images