Gas sensor performance degradation identification and early warning method

Abstract

This application relates to a gas sensor performance degradation identification and early warning method based on DTW and CUSUM in the field of sensor condition monitoring and fault diagnosis technology. The method includes the following steps: standardized calibration data acquisition and preprocessing; construction of a standard feature library; DTW degradation feature extraction; construction of a degradation time series; CUSUM mean shift detection; and degradation early warning and source tracing output. This invention integrates the DTW algorithm's ability to accurately characterize subtle differences in curve morphology with the CUSUM algorithm's advantage in accumulating and amplifying small gradual trends. It enables accurate identification and early warning of early, slow degradation in gas sensors, overcoming the technical shortcomings of traditional fixed threshold methods that rely solely on steady-state data, are insensitive to gradual degradation, and are prone to missed detections. This effectively improves the operational reliability and stability of coal mine safety monitoring systems and is applicable to preventative condition monitoring of various gas sensors in underground coal mines.

Description

Technical Field

[0001] This invention belongs to the field of sensor condition monitoring and fault diagnosis technology, specifically involving a method for identifying and warning of performance degradation of gas sensors based on Dynamic Time Warping (DTW) and Cumulative Sum (CUSUM). It is mainly applicable to preventive condition monitoring and proactive maintenance of various gas sensors in underground coal mine safety monitoring systems. Background Technology

[0002] Gas sensors such as methane, carbon monoxide, and oxygen are core sensing devices in coal mine safety monitoring systems. They directly undertake the task of real-time monitoring of underground environmental gas concentrations. The accuracy and timeliness of sensor measurement data directly affect the reliability of mine hazard identification and safety early warning.

[0003] Underground coal mines are characterized by high temperatures and humidity, large amounts of dust, and the presence of corrosive gases. Sensors operate under these harsh conditions for extended periods, causing their internal sensitive components and circuitry to age and deteriorate, gradually exhibiting performance degradation such as sensitivity decline, response hysteresis, zero-point drift, and distortion of the dynamic response curve. This type of degradation is a latent soft fault, not causing explicit malfunctions like power outages or no output, but it continuously leads to data distortion and inaccuracies. If the degradation trend is not identified in time, the prolonged operation of the sensor with these defects significantly increases the risk of missed or false alarms in mine monitoring.

[0004] Currently, the management of gas sensors in coal mines mainly relies on three methods: equipment self-diagnosis, periodic manual calibration, and mandatory scrapping upon expiration. Equipment self-diagnosis can only identify hard faults such as open circuits, short circuits, and complete component failure, but cannot identify slow performance degradation. Manual calibration is the primary means of judging sensor measurement accuracy on-site, involving regular testing of standard gas samples according to the "Coal Mine Safety Regulations," and determining compliance based on the deviation between the steady-state reading and the standard concentration. This detection method has significant technical defects: First, it only uses steady-state single-point data to evaluate the sensor's condition, ignoring the complete dynamic response process from gas introduction to numerical stability. Early sensor degradation often manifests as slower response, decreased upward slope, and slight curve distortion—dynamic characteristics where the steady-state error is still within acceptable limits, easily leading to missed detections. Second, the result of a single calibration is determined independently, failing to utilize data from the sensor's entire lifecycle, thus unable to capture long-term, slow, gradual degradation trends. It can only be passively detected after a fault becomes severe, lacking early warning capabilities.

[0005] To address the shortcomings of traditional detection methods, existing research has employed various algorithms for sensor state assessment; however, each individual algorithm has inherent limitations. Steady-state statistical methods, such as linear regression and the 3σ criterion, rely solely on steady-state numerical analysis and cannot utilize the detailed features of dynamic response curves, thus lacking the ability to identify weak and nonlinear early-stage degradation.

[0006] Machine learning methods such as SVM and neural networks rely on a large number of well-labeled fault samples. However, it is difficult to obtain batches of deteriorated samples with different aging levels at coal mine sites, resulting in limited model generalization ability. Furthermore, different types of sensors require retraining the model, making engineering implementation difficult.

[0007] Using curve matching methods such as Euclidean distance and single-time DTW alone can only achieve static comparison of a single curve, relying on fixed thresholds to judge anomalies. Downhole environmental noise and gas path fluctuations can easily cause fluctuations in single-time matching results, resulting in a high false alarm rate; at the same time, it is impossible to track the trend of long-term, small, gradual degradation, making it difficult to balance detection sensitivity and stability.

[0008] Existing technologies also include monitoring schemes that use the CUSUM algorithm alone, but these are only applicable to one-dimensional single-point time-series data, lack the ability to extract curve shapes in a refined manner, and cannot identify hidden degradation features such as dynamic response distortion. They are also slow to identify early performance degradation of sensors and have a high rate of missed detections.

[0009] In summary, existing technologies, whether relying on steady-state single-point data, performing only single curve comparisons, or conducting simple time-series statistics, cannot simultaneously achieve the extraction of subtle dynamic curve features and the cumulative detection of long-term weak degradation trends. Currently, there is no integrated degradation early warning scheme combining DTW curve morphology matching and CUSUM time-series cumulative detection. Therefore, there is an urgent need for an intelligent monitoring method that integrates the advantages of both types of algorithms to achieve accurate early identification and preventative early warning of gas sensor degradation. Summary of the Invention

[0010] This invention proposes a gas sensor performance degradation identification and early warning method based on DTW and CUSUM, aiming to solve the technical problems of existing coal mine gas sensor detection methods that rely on steady-state data, are insensitive to gradual degradation, have delayed early warning, and have a high rate of false positives and false negatives.

[0011] To address the aforementioned problems, this invention provides a method for identifying and warning of performance degradation in gas sensors based on DTW and CUSUM, including... S1 Standardized Calibration Data Acquisition and Preprocessing: Acquire the full response output sequence of the gas sensor from the introduction of standard gas sample to the stable response during the standardized calibration process of the automatic calibration device, and form the response curve time series for each calibration after data preprocessing; S2 Constructing a standard feature library: The standard feature library includes a normal class template curve generated by the dynamic time warping centroid averaging DBA algorithm from the response curves obtained by multiple normal gas sensors under the same standardized calibration process. S3 DTW Degradation Feature Extraction: For the gas sensor under test, calculate the dynamic time warping (DTW) distance between its current calibration response curve time series and the normal template curve, and extract degradation feature indices based on the optimal alignment path. The degradation feature indices are normalized DTW distances or statistics of the aligned residual sequence. S4 Constructing a degradation time series: Arrange the degradation characteristic indicators extracted from each calibration of the same gas sensor in chronological order to form a degradation time series of length N {X1, X2, …,Xn}; S5 CUSUM mean shift detection: The mean shift of the degraded time series is detected based on the cumulative sum CUSUM algorithm to obtain the CUSUM statistic and determine whether the gas sensor has experienced performance degradation. S6 Degradation Warning and Source Tracing Output: When the CUSUM statistic exceeds the preset control limit, it is determined that the gas sensor has degraded and a warning signal is issued.

[0012] To ensure the accuracy and validity of subsequent data, the standardized calibration process includes: introducing a standard gas sample of known concentration, keeping the inflation rate, flow rate, calibration duration, and environmental conditions constant; the preprocessing includes: removing peak interference at the moment of gas path switching, eliminating abnormal data that exceeds the sensor's measurement range, and smoothing filtering.

[0013] To obtain degradation characteristic indicators, this invention provides the following two methods: The first method is to calculate the optimal DTW cumulative distance D between the response curve of the gas sensor under test and the normal template curve, divide it by the alignment path length L, and obtain the normalized DTW distance D_norm = D / L. The normalized DTW distance is then used as the degradation characteristic index for this calibration.

[0014] The second method is to match the curve to be tested with the template curve point by point according to the DTW optimal alignment path to obtain the residual sequence, and calculate at least one of the standard deviation, root mean square value or kurtosis of the residual sequence as the degradation characteristic index of this calibration.

[0015] To ensure the accuracy of subsequent judgments, the specific process for mean offset detection based on the cumulative sum CUSUM algorithm is as follows: The target mean μ0 of the degradation time series is set as the mean value of the degradation index of the gas sensor under normal working conditions, which is calculated from the calibration data of the historical normal phase. Select the minimum allowable offset δ and the reference value K=δ / 2, as well as the control limit H; Calculate the one-sided upper offset cumulative sum statistic C + : C0 += 0 C_t + = max(0, C_{t-1} + + X_t-(μ0+ K)) When C_t + When H > H, it is determined that the degradation time series has a significant shift in mean at time t, that is, the performance of the gas sensor has deteriorated.

[0016] To provide data support for subsequent maintenance or replacement, the target mean μ0 is initially determined by the mean of the degradation characteristic index of the gas sensor during the first M normal calibrations after it is put into operation, and is re-initialized according to the subsequent maintenance or replacement situation after each performance degradation is determined.

[0017] To facilitate comparative early warning, when issuing an early warning signal, the approximate start time of the degradation is also output. The approximate start time is the corresponding calibration sequence in which the CUSUM statistic first leaves zero and continues to rise, which is found from the calibration records of the sensor.

[0018] To further improve the accuracy of the CUSUM algorithm, the standard feature library also includes response curves of multiple verified performance-degraded gas sensors under a standardized calibration process, as well as their corresponding degradation characteristic index sequences, to help determine the control limit H and reference value K of the CUSUM algorithm.

[0019] Preferably, the gas sensor is a methane sensor, carbon monoxide sensor, or oxygen sensor in a coal mine safety monitoring system.

[0020] To facilitate subsequent operation and maintenance guidance, a gas sensor performance degradation identification and early warning method is executed periodically after each automatic calibration to perform batch performance evaluations on all in-use gas sensors and generate a performance degradation early warning list.

[0021] In summary, this method combines the DTW algorithm's ability to finely match dynamic curve shapes with the CUSUM algorithm's ability to cumulatively detect subtle temporal shifts, thus overcoming the technical shortcomings of traditional detection methods. It can effectively identify progressive sensor performance degradation and provide early warning. Compared to existing technologies, it has the following significant advantages: (1) Make full use of dynamic response data throughout the entire process for more comprehensive detection dimensions: This invention relies on the DTW algorithm to perform matching analysis on the complete calibration response curve, which can effectively capture early degradation features that only exist in the dynamic process, such as response lag, slope decay, and local curve distortion, thus improving the problem of easy omission of latent faults in traditional steady-state single-point detection.

[0022] (2) Supports full life cycle trend analysis and has stronger detection continuity: This invention associates the historical calibration data of the sensor multiple times to construct the degradation time sequence, which changes the traditional detection mode of single independent judgment. It can continuously track the sensor performance change law and truly reflect the long-term degradation state of the equipment.

[0023] (3) Excellent ability to identify subtle and gradual degradation: With the help of the CUSUM cumulative amplification characteristic, the present invention can continuously accumulate small degradation shifts that are not significant in a single change, and can identify degradation trends in advance before the steady-state error of the sensor exceeds the tolerance, providing a basis for preventive maintenance.

[0024] (4) The algorithm has a high degree of adaptability and is applicable to a wider range of scenarios: The core parameters of this invention are adaptively tuned based on the normal historical data of the sensor itself, without the need for a lot of human experience intervention. It can adapt to gas sensors of different models, different service durations and different downhole working conditions, and has strong practicality.

[0025] (5) Strong operation and maintenance guidance: While outputting early warning results, this invention can trace the starting time of deterioration and quantify the degree of degradation, providing data support for on-site sensor calibration, maintenance and replacement, which can effectively reduce the situation of equipment operating with defects and improve the operation stability of the monitoring system.

[0026] (6) Dual-algorithm coupling and complementarity, breaking through the limitations of single algorithm: This invention integrates the technical advantages of DTW curve feature extraction and CUSUM time-series trend detection, which not only solves the problem that single DTW is difficult to identify long-term weak degradation trends, but also makes up for the deficiency of single CUSUM in being unable to mine the subtle morphological distortion of curves, significantly improving the overall accuracy and reliability of sensor degradation detection. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the gas sensor performance degradation identification and early warning method based on DTW and CUSUM of the present invention. Detailed Implementation

[0028] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0029] like Figure 1 The method for identifying and warning of performance degradation of a gas sensor based on DTW and CUSUM is shown. The gas sensor is a methane sensor, carbon monoxide sensor, or oxygen sensor in a coal mine safety monitoring system. The method for identifying and warning of performance degradation of the gas sensor includes the following steps: S1 Standardized Calibration Data Acquisition and Preprocessing The automatic calibration device obtains the full response output sequence of the gas sensor from the introduction of a standard gas sample to the stable response during the standardized calibration process. After data preprocessing, the response curve time series for each calibration is formed. The standardized calibration process includes: introducing a standard gas sample of known concentration, keeping the gas filling rate, flow rate, calibration duration, and environmental conditions constant; the preprocessing includes: removing peak interference at the moment of gas path switching, eliminating abnormal data that exceeds the sensor's measurement range, and smoothing filtering.

[0030] Specifically, the automatic calibration device performs periodic calibrations according to fixed operating conditions, maintaining consistency in standard gas concentration, inlet gas flow rate, calibration duration, and environmental parameters. It collects the complete output sequence of the gas sensor from baseline to stable response throughout the entire process. The raw data is preprocessed to remove gas path switching spikes and out-of-range anomalies, and random noise is eliminated through smoothing filtering to obtain a standardized response curve.

[0031] S2 builds a standard feature library The standard feature library includes a normal template curve generated by the dynamic time warping centroid averaging (DBA) algorithm from the response curves of multiple normal gas sensors obtained under the same standardized calibration process. It also includes the response curves of multiple confirmed performance-degraded gas sensors under the standardized calibration process, as well as their corresponding degradation feature index sequences, which are used to assist in determining the control limit H and reference value K of the CUSUM algorithm.

[0032] Specifically, multiple brand-new, metrologically qualified health gas sensors of the same model were selected, and multiple sets of normal response curves were collected under the same operating conditions. The DBA algorithm was used to iteratively average and generate a unique standard template curve as the benchmark for curve similarity comparison. At the same time, known deteriorated sample curves and corresponding degradation index sequences were included to assist in the tuning of CUSUM parameters.

[0033] S3 DTW Degradation Feature Extraction For the gas sensor under test, the dynamic time warping (DTW) distance between its current calibration response curve time series and the normal template curve is calculated, and degradation feature index is extracted based on the optimal alignment path. The degradation feature index is the normalized DTW distance or the statistics of the aligned residual sequence.

[0034] Specifically, the curve to be tested and the template curve are standardized, and the Sakoe-Chiba window constraint is introduced to solve for the optimal DTW alignment path. Based on the requirements, normalized DTW distance or residual statistical characteristics (standard deviation, root mean square, kurtosis) are selected as the degradation index for this calibration.

[0035] The specific degradation characteristic indicators are: (1) Calculate the optimal DTW cumulative distance D between the response curve of the gas sensor under test and the normal template curve, divide it by the alignment path length L, and obtain the normalized DTW distance D_norm = D / L. Use the normalized DTW distance as the degradation characteristic index of this calibration. (2) According to the optimal alignment path of DTW, the curve to be tested and the template curve are matched point by point to obtain the residual sequence. The standard deviation, root mean square value or kurtosis of the residual sequence are calculated as at least one of the degradation characteristic index of this calibration.

[0036] S4 Constructing Degradation Time Series The degradation characteristic indicators extracted from each calibration of the same gas sensor are arranged in chronological order to form a degradation time series of length N {X1, X2, …, Xn}. The degradation indicators from each calibration of the gas sensor are summarized and sorted by time to form a continuous degradation time series for subsequent trend analysis.

[0037] S5 CUSUM Mean Deviation Detection The mean shift of the degradation time series is detected based on the cumulative sum CUSUM algorithm to obtain the CUSUM statistic and determine whether the gas sensor has experienced performance degradation.

[0038] Specifically, the target mean μ0 is initialized using the normal data from the early stage of the gas sensor. The target mean μ0 is initially determined by the mean of the degradation characteristic index of the gas sensor during the first M normal calibrations after it was put into operation. After each performance degradation is determined, it is re-initialized according to the subsequent maintenance or replacement. The minimum degradation offset δ, reference value K and control limit H are set according to the accuracy requirements of on-site operation and maintenance. The one-sided cumulative sum statistic is calculated iteratively. When the statistic continuously exceeds the limit, it is determined that the sensor has experienced continuous performance degradation. The specific process for mean offset detection based on the cumulative sum CUSUM algorithm is as follows: The target mean μ0 of the degradation time series is set as the mean value of the degradation index of the gas sensor under normal working conditions, which is calculated from the calibration data of the historical normal phase. Select the minimum allowable offset δ and the reference value K=δ / 2, as well as the control limit H; Calculate the one-sided upper offset cumulative sum statistic C + : C0 + = 0 C_t + = max(0, C_{t-1} + + X_t-(μ0+ K)) When C_t + When H > H, it is determined that the degradation time series has a significant shift in mean at time t, that is, the performance of the gas sensor has deteriorated.

[0039] S6 Degradation Early Warning and Source Tracing Output When the CUSUM statistic exceeds the preset control limit, it is determined that the gas sensor has deteriorated and an early warning signal is issued. When the early warning signal is issued, the approximate start time of the deterioration is also output. The approximate start time is the corresponding calibration sequence in which the CUSUM statistic first leaves zero and continues to rise from the previous calibration records of the sensor.

[0040] Specifically, after an early warning is triggered, the sensor number, location, current degradation index, and warning time are output, and the statistical rise point is traced back to determine the approximate degradation start time. After equipment maintenance or replacement, the algorithm parameters are reset and the baseline mean is reinitialized to complete the status update.

[0041] In summary, this invention uses the DTW algorithm to precisely extract the dynamic response distortion features of the sensor and combines the CUSUM algorithm to amplify subtle degradation trends. This effectively improves the problems of early degradation being easily missed and the gradual trend recognition being lagging in traditional coal mine gas sensor detection methods. It can achieve early warning of the gradual performance degradation of the sensor and is suitable for long-term intelligent status monitoring and preventive operation and maintenance of various gas sensors in coal mines.

[0042] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape, principle and application direction of this application should be covered within the scope of protection of this application.

Claims

1. A method for identifying and warning of performance degradation of gas sensors based on DTW and CUSUM, characterized in that, include S1 Standardized Calibration Data Acquisition and Preprocessing: Acquire the full response output sequence of the gas sensor from the introduction of standard gas sample to the stable response during the standardized calibration process of the automatic calibration device, and form the response curve time series for each calibration after data preprocessing; S2 Constructing a standard feature library: The standard feature library includes a normal class template curve generated by the dynamic time warping centroid averaging DBA algorithm from the response curves obtained by multiple normal gas sensors under the same standardized calibration process. S3 DTW Degradation Feature Extraction: For the gas sensor under test, calculate the dynamic time warping (DTW) distance between its current calibration response curve time series and the normal template curve, and extract degradation feature indices based on the optimal alignment path. The degradation feature indices are normalized DTW distances or statistics of the aligned residual sequence. S4 Constructing a degradation time series: Arrange the degradation characteristic indicators extracted from each calibration of the same gas sensor in chronological order to form a degradation time series of length N {X1, X2, …,Xn}; S5 CUSUM mean shift detection: The mean shift of the degraded time series is detected based on the cumulative sum CUSUM algorithm to obtain the CUSUM statistic and determine whether the gas sensor has experienced performance degradation. S6 Degradation Warning and Source Tracing Output: When the CUSUM statistic exceeds the preset control limit, it is determined that the gas sensor has degraded and a warning signal is issued.

2. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 1, characterized in that, The standardized calibration process includes: introducing a standard gas sample of known concentration, keeping the inflation rate, flow rate, calibration duration, and environmental conditions constant; the preprocessing includes: removing spike interference at the moment of gas path switching, eliminating abnormal data that exceeds the sensor's measurement range, and smoothing filtering.

3. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 1, characterized in that, The degradation characteristic index is specifically as follows: calculate the optimal DTW cumulative distance D between the response curve of the gas sensor under test and the normal template curve, divide it by the alignment path length L to obtain the normalized DTW distance D_norm = D / L, and use the normalized DTW distance as the degradation characteristic index for this calibration.

4. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 1, characterized in that, The degradation characteristic index is specifically as follows: the test curve and the template curve are matched point by point according to the DTW optimal alignment path to obtain the residual sequence, and at least one of the standard deviation, root mean square value or kurtosis of the residual sequence is calculated as the degradation characteristic index of this calibration.

5. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 1, characterized in that, The specific process of mean offset detection based on the cumulative sum and CUSUM algorithm is as follows: The target mean μ0 of the degradation time series is set as the mean value of the degradation index of the gas sensor under normal working conditions, which is calculated from the calibration data of the historical normal phase. Select the minimum allowable offset δ and the reference value K=δ / 2, as well as the control limit H; Calculate the one-sided upper offset cumulative sum statistic C + : C0 + = 0 C_t + = max(0, C_{t-1} + + X_t-(μ0+ K)) When C_t + When H > H, it is determined that the degradation time series has a significant shift in mean at time t, that is, the performance of the gas sensor has deteriorated.

6. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 5, characterized in that, The target mean μ0 is initially determined by the mean of the degradation characteristic index of the gas sensor during the first M normal calibrations after it is put into operation, and is re-initialized according to the subsequent maintenance or replacement situation after each performance degradation is determined.

7. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 1, characterized in that, When issuing a warning signal, the approximate start time of the degradation is also output. The approximate start time is the corresponding calibration sequence in which the CUSUM statistic first leaves zero and continues to rise, which is found from the calibration records of the sensor.

8. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 1, characterized in that, The standard feature library also includes response curves of multiple confirmed performance-degraded gas sensors under a standardized calibration process, as well as their corresponding degradation feature index sequences, to help determine the control limit H and reference value K of the CUSUM algorithm.

9. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 1, characterized in that, The gas sensor is a methane sensor, carbon monoxide sensor, or oxygen sensor in a coal mine safety monitoring system.

10. The gas sensor performance degradation identification and early warning method based on DTW and CUSUM according to claim 1, characterized in that, After each automatic calibration, a gas sensor performance degradation identification and early warning method is executed periodically to perform batch performance evaluations on all gas sensors in use and generate a performance degradation early warning list.

Images