A sensor state diagnosis method and device, electronic equipment, and storage medium

Abstract

The application discloses a sensor state diagnosis method and device, an electronic device and a storage medium. The method comprises the following steps: inputting environment data of an environment in which a sensor is located and a detection value of a health index of the sensor into a health index prediction model to obtain prediction values of various health indexes of the healthy sensor and the sensor in the same environment; and determining a health state of the sensor according to a difference between the detection value and the prediction value. According to the diagnosis method provided by the application, the health index prediction model with a deep self-encoder architecture is used to predict the value of the health index of the sensor in the healthy state, and the prediction value is obtained. The high-order features reflecting the running state of the sensor can be adaptively extracted from the environment data and the multi-dimensional health index detection value. The health state of the sensor is determined according to the difference between the detection value and the prediction value. The weak drift signal can be effectively amplified and captured in the residual sequence, and therefore the accuracy of diagnosis is improved.

Description

Technical Field

[0001] This application belongs to the field of sensor monitoring technology, specifically relating to a sensor status diagnosis method and device, electronic equipment, and storage medium. Background Technology

[0002] In underground mining operations, multi-parameter gas sensors used to monitor various toxic and harmful gases such as CO, H2S, NO2, and SO2 are core sensing components for safety assurance. Due to the long-term exposure to extreme conditions in underground environments, such as high humidity, high dust concentrations, strong electromagnetic interference, and large temperature variations, sensors commonly suffer from performance degradation, leading to inaccurate monitoring data and impacting the accuracy of safety decisions.

[0003] Currently popular sensors do not have online self-diagnosis capabilities for their health status. Common performance degradation issues such as zero-point drift, sensitivity decay, and response hysteresis can only be detected after periodic calibration or when a fault occurs. The diagnosis is highly delayed and lacks real-time capability.

[0004] Although some intelligent diagnostic methods that incorporate neural networks have certain online real-time diagnostic capabilities, these methods rely on manually constructed feature samples and are difficult to fully explore the deep nonlinear coupling relationship between multi-dimensional environmental parameters such as temperature, humidity, dust, and electromagnetic interference and sensor performance degradation. They are particularly insensitive to early weak drift signals and have a high false negative rate.

[0005] Therefore, there is an urgent need for a sensor condition diagnosis method that balances accuracy and real-time performance. Summary of the Invention

[0006] This application provides a sensor status diagnosis method and device, electronic device, and storage medium, which can solve the problems of weak real-time performance or poor accuracy of traditional diagnostic methods.

[0007] In a first aspect, embodiments of this application provide a sensor state diagnosis method, the method comprising: Acquire environmental data and sensor readings of various health indicators; The environmental data and the detection value are input into the health indicator prediction model to obtain the predicted value of the health indicator. The health indicator prediction model adopts a deep autoencoder architecture. The predicted value represents the value of the health indicator when the healthy sensor and the sensor are in the same operating environment. The health status of the sensor is determined based on the difference between the detected value and the predicted value.

[0008] In one possible implementation of the first aspect, determining the health status of the sensor based on the difference between the detected value and the predicted value includes: The reconstruction error is determined based on the difference between the detected value and the predicted value; By comparing the reconstruction error with the current fault threshold, the health status of the sensor is determined based on the degree to which the reconstruction error deviates from the current fault threshold; the health status includes healthy, unhealthy, and faulty.

[0009] In one possible implementation of the first aspect, before determining the health status of the sensor based on the difference between the detected value and the predicted value, the method further includes: The current fault threshold is determined based on the distribution of sample reconstruction errors in the previous training cycle.

[0010] In one possible implementation of the first aspect, the current fault threshold is equal to the sum of the mean and multiple standard deviations of the sample reconstruction error.

[0011] In one possible implementation of the first aspect, when the sensor is in an unhealthy state, the method further includes: The environmental data and the real-time output value of the sensor are input into the measurement deviation prediction model to obtain the measurement deviation of the sensor; The real-time output value is corrected using the measurement deviation to obtain the initial calibrated output value of the sensor. The output value after the initial correction is corrected using a cross-interference matrix to obtain the corrected output value of the sensor. The cross-interference matrix is ​​used to characterize the effect of cross-interference between sensors.

[0012] In one possible implementation of the first aspect, the bottleneck layer of the deep autoencoder includes neurons that correspond one-to-one with the types of the health indicators; for each neuron, the process of configuring the activation function on the neuron includes: The activation function of the neuron is configured according to the value range of the corresponding health indicators of the neuron.

[0013] In one possible implementation of the first aspect, the method further includes: The detected value is input into the life prediction model to obtain the remaining lifespan of the sensor; If the remaining service life is less than the alarm threshold, an alarm notification will be sent to the user.

[0014] Secondly, embodiments of this application provide a sensor state diagnostic device, including an acquisition unit, a prediction unit, and a diagnostic unit; The acquisition unit is used to acquire environmental data and the detection values ​​of multidimensional health indicators from sensors; The prediction unit is used to input the environmental data and the detection value into the health indicator prediction model to obtain the predicted value of the health indicator. The health indicator prediction model adopts a deep autoencoder architecture, and the predicted value represents the value of the health indicator when the healthy sensor and the sensor are in the same operating environment. The diagnostic unit is used to determine the health status of the sensor based on the difference between the detected value and the predicted value.

[0015] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory, wherein the memory is used to store a computer program; the processor can be used to execute a calculator program (instructions) stored in the memory to implement the method described in the first aspect.

[0016] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed, can implement the method described in the first aspect above.

[0017] The beneficial effects of this application embodiment compared with the prior art are as follows: Since the deep autoencoder learns the data distribution pattern under normal conditions through nonlinear mapping, its multi-layer latent variable structure naturally possesses the ability to model the deep nonlinear coupling relationship between complex environmental factors such as temperature, humidity, and electromagnetic interference and sensor performance degradation. The diagnostic method provided in this application, by adopting a health indicator prediction model with a deep autoencoder architecture, inputs environmental data and the detection values ​​of health indicators into the health indicator prediction model to predict the value of the sensor's health indicators under healthy conditions, and obtains the predicted value. It can adaptively extract high-order features reflecting the sensor's operating status from environmental data and multi-dimensional health indicator detection values; and determine the sensor's health status based on the difference between the detected value and the predicted value. It can effectively amplify and capture weak drift signals in the residual sequence, thereby improving the accuracy of diagnosis. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the implementation of a sensor management method provided in this application embodiment; Figure 2 A flowchart illustrating the implementation of a sensor state diagnosis method provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of a health indicator prediction model provided in an embodiment of this application; Figure 4 A flowchart illustrating the implementation of a sensor calibration method provided in this application embodiment; Figure 5 This application provides a sensor health status diagnostic device. Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0020] It should be understood that, when used in this application specification, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0021] It should also be understood that the term “and / or” as used in this application specification means any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.

[0022] As used in this application specification, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [the described condition or event] is detected," or "in response to detection of [the described condition or event]."

[0023] Furthermore, in the description of this application, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0024] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0025] The present application will be described in further detail below with reference to specific embodiments, but the implementation of the present application is not limited thereto.

[0026] The method provided in this application can be applied to electronic devices such as mobile terminals, personal laptops, and supercomputers. This application does not impose any restrictions on the specific type of electronic device.

[0027] Example 1 Figure 1 The flowchart shown is an implementation flowchart of a sensor management method provided in an embodiment of this application. It is an example and not a limitation. The method may include steps S101-S104, and each step is described below.

[0028] S101 acquires environmental data, sensor readings of various health indicators, and real-time sensor output values.

[0029] For example, environmental data may include temperature, relative humidity, dust concentration, electromagnetic interference intensity, etc.

[0030] For example, the health indicators of a sensor may include zero-point offset, sensitivity coefficient, response time, noise amplitude, and baseline stability.

[0031] For example, the real-time output value of the sensor may include one or more of the following: the concentration of toxic and harmful gases such as CO, H2S, NO2, and SO2.

[0032] Optionally, after acquiring a type of data, preprocessing operations such as filtering, denoising, and time alignment can be performed on various types of data within that type of data to eliminate the effects of random noise and transmission delay, and ensure the time consistency of data from each channel.

[0033] For example, after acquiring environmental data, different types of data such as temperature, relative humidity, dust concentration, and electromagnetic interference intensity can be preprocessed. Alternatively, environmental data, health indicator detection values, and real-time sensor output values ​​can be preprocessed together.

[0034] It should be understood that there is at least one sensor here, and typically multiple sensors in a sensor array. Furthermore, the sensors operate in harsh environments, typically underground mines.

[0035] For example, consider a large underground metal mine with an annual output of 1.5 million tons. The underground operating areas include three sections: -100m, -200m, and -300m. The underground environment is harsh: relative humidity remains between 85% and 95% year-round, dust concentration is 50–150 mg / m³, and there are strong electromagnetic interference sources such as frequency converters and large motors. The main toxic and harmful gases underground include CO (generated by explosions), H₂S (generated by ore oxidation), NO₂ (generated by explosions), and SO₂ (generated by the oxidation of sulfur-containing ores). Therefore, the sensors to be diagnosed include 80 multi-parameter gas sensors (integrating four electrochemical detection elements: CO, H₂S, NO₂, and SO₂) deployed at key locations such as the mining faces, transport roadways, and edges of goaf areas in each section. Each sensor has a built-in temperature sensor, humidity sensor, dust concentration detection module, and electromagnetic interference detection module, collecting environmental data in real time at a fixed sampling frequency, such as 1Hz. The sensor can upload the data it collects to ground-based electronic equipment (equivalent to a ground station) via industrial Ethernet. Similarly, it can upload at a fixed frequency, such as 10Hz. Accordingly, the electronic equipment obtains environmental data, the sensor's various health indicator detection values, and the sensor's real-time output values.

[0036] S102, determine the health status of the sensor based on environmental data and the detected values ​​of health indicators.

[0037] In one example, the health status of a sensor can be categorized into three types: healthy, unhealthy, and faulty.

[0038] For example, when the sensor is in a healthy state, it means that all indicators of the sensor are normal, the monitoring accuracy meets the requirements, no intervention is needed, and the real-time output value of the sensor can be directly used as the final output value.

[0039] For example, when the sensor is in an unhealthy state, it means that the sensor's indicators have deviated. However, this deviation can be corrected. The following step S103 can be performed to correct the real-time output value of the sensor and use the corrected output value as the final output value.

[0040] Specifically, sensor unhealthy states can be categorized into two types: sub-healthy and abnormal. In a sub-healthy state, it indicates that only some sensor metrics deviate, or the deviations are minor. While correcting these deviations, push notifications can be sent to the client software used by staff to remind them to pay attention to the sensor. In an abnormal state, it indicates that the sensor metrics deviate significantly. Although this can be mitigated through correction, staff need to perform maintenance and adjustments as soon as possible. In this case, maintenance tasks can be issued to the client software used by staff to remind them to maintain the sensor promptly.

[0041] For example, when the health status of a sensor is in a faulty state, it means that the sensor's indicators are seriously deviated or have failed. Simply correcting it cannot improve its accuracy to a tolerable range, and the sensor needs to be replaced immediately.

[0042] S103, corrects the real-time output value of the sensor based on environmental data to obtain the corrected output value.

[0043] In one example, the real-time output value of the sensor can be initially corrected based on the measurement deviation of the sensor predicted by the model, and then the output value after the initial correction can be corrected a second time by the cross-interference matrix between the sensors to obtain the corrected output value.

[0044] For example, the cross-interference matrix is ​​used to characterize the effect of cross-interference between sensors.

[0045] S104 predicts the remaining lifespan of the sensor based on the detected values ​​of health indicators.

[0046] For example, step S104 can be performed after step S102. This application does not limit the order of steps S104 and S103.

[0047] In one example, a lifetime prediction model can be built based on the changing trajectories of various health indicators in a sample sensor over time. The lifetime prediction model can use a combination of exponential smoothing and linear regression to predict the time when various health indicators of the sensor reach the failure threshold, and the shortest one is taken as the remaining lifetime of the sensor.

[0048] For example, the sample sensor can be a previously used sensor.

[0049] Specifically, a unique digital identity can be established for each sample sensor, recording all operational data, health indicator values, compensation parameters, and abnormal events from the time the sensor is put into use until it is discontinued. Then, the lifespan utilization model can be trained using the trajectory of various health indicators over time recorded in the health record as sample data.

[0050] For example, taking a CO sensor channel in the -200m mid-section as an example, since its commissioning in January 2025, the trajectory of its health indicators has been continuously recorded. Taking January, April, July, and October (current time) as nodes, the trajectory of zero-point offset change is: 0.88, 1.5, 2.8, 4.2; the trajectory of sensitivity coefficient change is: 0.98, 0.94, 0.87, 0.82; and the trajectory of response time change is: 18, 22, 28, 35. Based on the above trajectories, the lifetime prediction network predicts that the sensor's zero-point offset will fail in March 2026 (Z≥8ppm), the sensitivity coefficient will fail in February 2026 (S≤0.70), and the response time will fail in April 2026 (τ≥45s). The model can output that the sensor's remaining lifetime is 4 months, or it can output that the sensor's remaining lifetime is between 4 and 6 months.

[0051] For example, the compensation parameters may include one or more of the following: the difference between the corrected output value and the real-time output value of the sensor, the difference between the output value after initial correction and the real-time output value, and the cross-interference matrix.

[0052] Optionally, when it is predicted that the remaining lifespan of the sensor is below the alarm threshold, maintenance reminders and replacement suggestions can be automatically generated and pushed to the client of relevant personnel.

[0053] For example, the electronic device can push a message to the client of the relevant staff: "The comprehensive health score of the -200m mid-section CO sensor is 68 points (out of 100), with an estimated remaining service life of 5 months. It is recommended to arrange replacement before February 2026."

[0054] The management method provided in this application can determine the current health status of the sensor in real time by integrating environmental data and health indicator detection; it can ensure measurement accuracy by correcting the sensor output value using an environmental compensation mechanism; it can provide a basis for preventive maintenance of the sensor by predicting the remaining lifespan of the sensor based on the degradation trend of health indicators; and it can achieve full-cycle health management of the sensor through a management chain from state perception to accuracy assurance to lifespan prediction, thereby improving the reliability of the sensor, reducing the risk of sudden failures, and extending the service life of the sensor.

[0055] Example 2 Figure 2 This is a flowchart illustrating the implementation of a sensor state diagnosis method according to an embodiment of this application. As an example and not a limitation, method 200 is a specific possible implementation of steps S101 and S102 in method 100 described above. Method 200 includes steps S201-S203, which are described below.

[0056] S201 acquires environmental data and sensor readings of various health indicators.

[0057] For example, the method for obtaining environmental data and health indicator detection values ​​is the same as in step S101 above. Please refer to the relevant description in step S101 for details, which will not be repeated here.

[0058] S202, input the environmental data and the detected values ​​of health indicators into the health indicator prediction model to obtain the predicted values ​​of health indicators.

[0059] For example, the predicted value represents the value of a health indicator when the healthy sensor and the sensor to be diagnosed are in the same operating environment.

[0060] In some embodiments, the health indicator prediction model can employ a deep autoencoder architecture, comprising an encoder and a decoder. The encoder can progressively compress high-dimensional input features to a low-dimensional bottleneck layer through a multi-layer fully connected network, and the decoder then reconstructs the low-dimensional features back to the original dimensions to obtain the predicted values ​​of the health indicators.

[0061] For example, environmental data and health indicators can be combined to form a high-dimensional feature vector F=[Z,S,τ,N,σ,T,RH,D,EMI] and input into the health indicator prediction model. Here, Z, S, τ, N, and σ are the detected values ​​of zero offset, sensitivity coefficient, response time, noise amplitude, and baseline stability, respectively, and T, RH, D, and EMI represent temperature, relative humidity, dust concentration, and electromagnetic interference intensity, respectively.

[0062] For example, see Figure 3 The encoder can adopt a three-layer cascaded network structure with 9, 6, and 3 neurons in each layer, and the decoder can also adopt a three-layer cascaded network structure with 3, 6, and 9 neurons in each layer.

[0063] In one possible implementation, during the initial calibration and break-in phase of the sensor's initial use, at least 168 hours of normal operation data (i.e., the detection values ​​of health indicators) can be collected. Then, a benchmark dataset for each detection channel (i.e., different types of health indicators) can be constructed. Based on the benchmark dataset, the health indicator prediction model can be trained unsupervised, enabling the health indicator prediction model to accurately memorize and reconstruct the normal mode of the sensor under healthy operating conditions.

[0064] For example, after training is completed, the baseline values ​​H0=[Z0,S0,τ0,N0,σ0] of each detection channel can be obtained based on the baseline dataset, which can be used as a reference standard for subsequent health status assessment.

[0065] For example, taking the CO sensor channel of a mining face at -200m as an example, we can first collect 168 hours of operating data and environmental data of the sensor to construct a benchmark dataset including 168 hours × 3600 sampling points ≈ 604,800 samples. Then, we train a health index prediction model to obtain the benchmark values ​​H0=[0.8,0.98,18,0.3,0.12] for each detection channel. Among them, the zero offset Z0=0.8ppm (after zero-point calibration with standard gas), the sensitivity coefficient S0=0.98 (50ppm response value of standard gas / standard value), the response time τ0=18s (time from contact with standard gas to reaching 90% of the stable value), the noise amplitude N0=0.3ppm (peak value), and the baseline stability σ0=0.12ppm (24h baseline standard deviation).

[0066] For example, during training, a loss function based on mean squared error can be used.

[0067] Optionally, after the health indicator prediction model starts being used, relevant data for each model prediction can be recorded synchronously, such as input environmental data, detected values ​​of health indicators, and output predicted values, as a running dataset. If the timing / prediction accuracy deteriorates, samples can be synchronously sampled from both the baseline dataset and the running dataset for retraining.

[0068] In one example, to incorporate the physical laws governing sensor degradation into the prediction process of a health indicator prediction model, neurons corresponding one-to-one with the types of health indicators can be set in the bottleneck layer of the model. The activation function of each neuron is configured based on the value range of the corresponding type of health indicator.

[0069] For example, see Figure 3 In the last layer of the encoder, a fully connected layer consisting of these 5 neurons can be connected as a bottleneck layer, and the input of the bottleneck layer is the output of the decoder.

[0070] For example, see Figure 3 Neurons 1-5 correspond to five health indicators: zero-point offset, sensitivity coefficient, response time, noise amplitude, and baseline stability. Neurons 1 and 4 can use linear activation functions, allowing their values ​​to fluctuate within a positive and negative range; while sensitivity coefficient, response time, and baseline stability are usually non-linear and can use the Softplus activation function (i.e.,...). Ensure the output is non-negative.

[0071] In another example, the physical laws governing sensor degradation can be incorporated into the prediction process of a health indicator prediction model by adding a physical constraint loss to the loss function.

[0072] For example, taking advantage of the monotonically non-increasing nature of the sensitivity coefficient, a loss function term can be designed, such as... This forces the indicator prediction model to learn the physical law that the sensitivity will only decrease or remain unchanged.

[0073] For example, considering that the trend of zero-point offset generally satisfies the Arrhenius equation, a loss function term can be designed to calculate the difference between the predicted value of zero-point offset output by the index prediction model and the separately calculated predicted value that satisfies the Arrhenius equation. The loss function term is then used to penalize the value of the larger difference, thereby forcing the index prediction model to learn the physical laws satisfied by the zero-point offset.

[0074] S203, determine the health status of the sensor based on the difference between the detected value and the predicted value of the health indicator.

[0075] In one possible implementation, the reconstruction error can be determined first based on the difference between the detected and predicted values ​​of the health indicator; then, the reconstruction error can be compared with the current fault threshold, and the health status of the sensor can be determined based on the degree to which the reconstruction error deviates from the current fault threshold.

[0076] In one example, the reconstruction error can be the square of the L2 norm of the difference between the detected and predicted values ​​of a health indicator. The larger the value, the more serious the deviation of the sensor's current state from the health benchmark.

[0077] For example, the reconstruction error can satisfy the following formula:

[0078] in, For reconstruction error, , These are the measured and predicted values ​​of health indicators, respectively.

[0079] For example, a target multiple can be set. For instance, if the reconstruction error is less than 1 times the fault threshold, it is determined to be in a healthy state; if it is between 1 and 3 times, it is in an unhealthy state; and if it is greater than 3 times, it is in a fault state.

[0080] For example, taking the current fault threshold of 0.0023 as an example, if an abnormality is detected in the operating data of a CO sensor on a certain day in May 2026, and the real-time feature vectors of 60 consecutive sampling points are input into a trained health indicator prediction model, the calculated reconstruction error RE = 0.0187 is far exceeding 3 times the threshold, then it can be determined that the sensor is in a faulty state.

[0081] In one example, the current fault threshold can be dynamically adjusted based on the distribution of reconstruction error in the previous training cycle.

[0082] For example, the current fault threshold can be the sum of the mean and multiple standard deviations of the reconstruction error in the previous training cycle, such as the sum of the mean and three standard deviations.

[0083] Compared to some traditional diagnostic methods that use Long Short-Term Memory (LSTM) networks to directly predict health indicators based on sensor output values ​​and then determine sensor status based on the predicted health indicators, the diagnostic method provided in this application is based on deep autoencoders to predict values. It uses residual analysis of the difference between the detected and predicted values ​​to determine the sensor status. This method can leverage the characteristics of deep autoencoders, which are good at reconstructing normal pattern data and whose reconstruction errors are more sensitive to anomalies in abrupt and gradual changes, to improve the detection rate of unknown faults. Furthermore, it does not require a pre-set degradation model, thus reducing the implementation difficulty.

[0084] Optionally, when it is determined that the sensor is in an abnormal or faulty state, method 200 may further include step S204.

[0085] S204. Determine the anomaly type based on the distribution of reconstruction errors.

[0086] For example, if the zero-point offset contributes the most to the reconstruction error, the sensor's anomaly / fault can be determined to be a zero-point abrupt change; if the sensitivity coefficient contributes the most, the sensor's anomaly / fault can be determined to be a sensitivity drop; if the response time contributes the most, the sensor's anomaly / fault can be determined to be a response hysteresis; if the noise amplitude contributes the most, the sensor's anomaly / fault can be determined to be a noise aberration; if the residuals of multiple health indicators (i.e., the difference between the predicted and detected values) increase simultaneously, it is a composite type.

[0087] For example, continuing with the above example, the contributions of the various health indicators of the sensor to the reconstruction error are as follows: the contribution rate of the zero-point offset Z-dimensional residual is 62%, the contribution rate of the sensitivity coefficient S-dimensional residual is 28%, and the remaining dimensions total 10%. It can be determined that the anomaly type of the sensor is a composite anomaly of "primarily zero-point drift and secondarily sensitivity attenuation", and this anomaly information will be pushed to the client of the relevant personnel.

[0088] Since deep autoencoders learn the data distribution patterns under normal conditions through nonlinear mapping, their multi-layered latent variable structure inherently possesses the ability to model the deep nonlinear coupling relationship between complex environmental factors such as temperature, humidity, and electromagnetic interference and sensor performance degradation. The diagnostic method provided in this application, by employing a health indicator prediction model based on a deep autoencoder architecture, inputs environmental data and the detected values ​​of health indicators into the health indicator prediction model to predict the values ​​of the sensor's health indicators under healthy conditions, obtaining the predicted values. It can adaptively extract high-order features reflecting the sensor's operating status from environmental data and multi-dimensional health indicator detection values; and determine the sensor's health status based on the difference between the detected values ​​and the predicted values. It can effectively amplify and capture weak drift signals in the residual sequence, thereby improving the accuracy of diagnosis.

[0089] Furthermore, by reconstructing the distribution of errors to locate the anomaly type, the shortcomings of traditional fault diagnosis, which only knows the anomaly but not the cause, can be overcome, thus improving the interpretability of the diagnostic results.

[0090] Example 3 Figure 4 This is a flowchart illustrating the implementation of a sensor calibration method according to an embodiment of this application. As an example and not a limitation, method 400 is a specific possible implementation of step S103 described above. Method 400 may include steps S401-S403, which are described below.

[0091] S401 inputs environmental data and the real-time output value of the sensor into the measurement deviation prediction model to obtain the sensor's measurement deviation.

[0092] For example, a measurement bias prediction model can employ an LSTM architecture.

[0093] Specifically, the measurement bias prediction model based on the LSTM architecture can dynamically learn the nonlinear mapping relationship between environmental data and measurement bias through a gating mechanism, allowing the cell state to retain long-term memory of environmental change patterns and output accurate prediction compensation (i.e., measurement bias).

[0094] For example, in July 2026, the mine entered the rainy season, and the relative humidity underground continuously rose to 93%–97%. A persistent positive deviation was detected in the measurement value of an H2S sensor channel in the -100m section. The input environmental data was "T=32℃, RH=95%, D=85mg / m³, EMI=2.3V / m (power frequency interference)". The real-time output value of the H2S was " =8.2ppm”, these data are input into the measurement deviation prediction model, and the model outputs the measurement deviation as “ =1.8ppm (positive deviation, mainly due to high humidity).

[0095] Optionally, the measurement bias prediction model can use a sliding time window for online incremental learning to achieve continuous adaptive optimization of model parameters as the environment changes.

[0096] For example, the width of the sliding time window can be 60 sampling points, and the step size can be 10 sampling points.

[0097] S402 uses the real-time output value of the measurement deviation correction sensor to obtain the output value after the initial correction.

[0098] In one example, the measurement deviation can be used to correct the real-time output value in real time, resulting in the output value after initial correction.

[0099] For example, the output value after initial correction can satisfy the following formula:

[0100] in, For the sensor The output value after the initial correction at time . For the sensor Real-time output value at any given moment. For the sensor Measurement deviation at time.

[0101] For example, referring to the example above, the sensor's real-time output value for H2S is " =8.2ppm”, the model output measurement deviation is “ =1.8ppm", then the output value after the initial calibration is " =8.2-1.8=6.4ppm. After on-site verification, the actual concentration of H2S was 6.5ppm, and the error after correction was only 0.1ppm (1.5%), which was far lower than the measurement error of 1.7ppm (26%) before correction.

[0102] S403 uses the cross-interference matrix to correct the output value after the initial correction, and obtains the corrected output value of the sensor.

[0103] For example, the cross-interference matrix is ​​used to characterize the impact of cross-interference between sensors on sensor measurements.

[0104] In one possible implementation, a cross-interference matrix needs to be constructed before performing secondary correction.

[0105] For example, the cross-interference matrix can be calibrated offline in the laboratory before the sensor is enabled, and then corrected online based on the sensor's usage data after the sensor is enabled.

[0106] In one example, during offline calibration, multiple sets of mixed standard gases with known concentrations can be configured, including sensor detection objects such as CO, H2S, NO2, and SO2. Then, the raw sensor readings are recorded, and based on the solution function of the cross-interference matrix, each element in the cross-interference matrix is ​​fitted by the least squares method according to the known true concentration and raw reading of each gas set.

[0107] For example, the function for solving the cross-interference matrix satisfies the following formula:

[0108] in, , , , These represent the interference deviations of the sensor's real-time output values ​​for CO, H2S, NO2, and SO2, respectively. For the first The gas for the first The cross-interference coefficient of the gas sensor , , , This refers to the real-time output values ​​of the sensor for CO, H2S, NO2, and SO2.

[0109] In one example, during the online correction process, the readings of a high-precision downhole sensor with minimal error can be used as the true gas concentration to infer the deviation of the current cross-interference matrix. A Bayesian optimization algorithm is then used to perform a small-scale search around the initial cross-interference matrix (i.e., the cross-interference matrix determined during offline calibration) to find a new matrix that makes the corrected output value closest to the true concentration.

[0110] Optionally, online correction operations can be performed periodically to update the cross-interference matrix regularly and maintain correction accuracy.

[0111] In one possible implementation, the interference deviation can be determined using the cross-interference matrix and the real-time output value of the sensor. Then, the interference deviation is superimposed on the output value after the initial correction to obtain the corrected output value.

[0112] For example, on a certain day in September 2026, two gases, CO (due to poor ventilation after an explosion) and SO2 (due to oxidation of sulfur-containing ore), were simultaneously present in a certain area at a depth of -300m. The sensor's real-time output value for SO2 was "12ppm," while the actual concentration of CO was "35ppm." The interference coefficient m of CO on SO2 was... 41=0.08 (meaning that every 1 ppm of CO causes 0.08 ppm of interference to the SO2 sensor), the calculated interference deviation is "35 × 0.08 = 2.8 ppm". Previously, the sensor's measurement deviation for SO2 was calculated to be "-0.1 ppm". Therefore, based on the initial calibrated output value of "12.1 ppm", the interference deviation is added, resulting in a final corrected output value of "12.1 - 2.8 = 9.3 ppm". Subsequent field testing showed that the actual SO2 concentration in the environment was 9.5 ppm. This indicates that the error after secondary calibration is less than that after primary calibration, and secondary calibration using a cross-interference matrix can further improve monitoring accuracy.

[0113] The calibration method provided in this application performs two-stage calibration of the sensor output value through a deviation prediction model and a cross-interference matrix. The initial calibration effectively eliminates errors introduced by environmental factors, while the secondary calibration eliminates the cross-coupling effects between sensor arrays. This overcomes the measurement deviation problem caused by insufficient selectivity of electrochemical sensors in environments with multiple coexisting gases, thereby improving the sensor's output accuracy. Furthermore, for the cross-interference matrix, an offline calibration combined with a Bayesian optimization online correction strategy is adopted, which can significantly improve the measurement accuracy and long-term stability of the multi-sensor system while ensuring real-time response.

[0114] Example 4 Figure 5 The diagram shown illustrates the structure of a sensor health status diagnostic device according to an embodiment of this application. As an example and not a limitation, the device 500 may include an acquisition unit 510, a prediction unit 520, and a diagnostic unit 530.

[0115] In some embodiments, the acquisition unit 510 is used to acquire environmental data and the detection values ​​of multidimensional health indicators of the sensor; the prediction unit 520 is used to input the environmental data and the detection values ​​into the health indicator prediction model to obtain the predicted values ​​of the health indicators; and the diagnosis unit 530 is used to determine the health status of the sensor based on the difference between the detection values ​​and the predicted values.

[0116] For example, the health indicator prediction model employs a deep autoencoder architecture.

[0117] Optionally, the bottleneck layer of the deep autoencoder includes neurons that correspond one-to-one with the types of health indicators, and the activation functions on these neurons are configured according to the value range of their respective health indicator types.

[0118] For example, the predicted value represents the health index value of a healthy sensor and the sensor when it is in the same operating environment.

[0119] In one possible implementation, the diagnostic unit 530 is specifically used to: determine a reconstruction error based on the difference between the detected value and the predicted value; compare the reconstruction error with the current fault threshold, and determine the health status of the sensor based on the degree to which the reconstruction error deviates from the current fault threshold.

[0120] For example, health status includes healthy, unhealthy, and faulty.

[0121] In one possible implementation, before the diagnostic unit 530 determines the health status, the diagnostic unit 530 is also used to determine the current fault threshold based on the distribution of sample reconstruction errors in the previous training cycle.

[0122] In one example, the current fault threshold is equal to the sum of the mean and multiple standard deviations of the sample reconstruction error.

[0123] In one possible implementation, the device 500 may further include a correction unit 540, configured to: input environmental data and the real-time output value of the sensor into a measurement deviation prediction model to obtain the measurement deviation of the sensor; correct the real-time output value using the measurement deviation to obtain the output value of the sensor after initial correction; and correct the output value of the sensor after initial correction using a cross-interference matrix to obtain the corrected output value of the sensor, wherein the cross-interference matrix is ​​used to characterize the influence of cross-interference between sensors.

[0124] In one possible implementation, the device 500 may further include a lifespan prediction unit 550, which inputs health indicators into a lifespan prediction model to obtain the remaining lifespan of the sensor; and sends an alarm notification to the user if the remaining lifespan is less than an alarm threshold.

[0125] Because deep autoencoders learn the data distribution patterns under normal conditions through nonlinear mapping, their multi-layered latent variable structure inherently possesses the ability to model the deep nonlinear coupling relationship between complex environmental factors such as temperature, humidity, and electromagnetic interference and sensor performance degradation. The device provided in this application, by employing a health indicator prediction model based on a deep autoencoder architecture, inputs environmental data and detected values ​​of health indicators into the health indicator prediction model to predict the values ​​of the sensor's health indicators under healthy conditions, obtaining predicted values. It can adaptively extract high-order features reflecting the sensor's operating status from environmental data and multi-dimensional health indicator detection values; and determine the sensor's health status based on the difference between the detected values ​​and the predicted values. It can effectively amplify and capture weak drift signals in the residual sequence, thereby improving the accuracy of diagnosis.

[0126] Furthermore, the sensor output value is corrected in two stages by using a deviation prediction model and a cross-interference matrix. The initial correction can effectively eliminate errors introduced by environmental factors, while the secondary correction can eliminate the cross-coupling effect between sensor arrays. This overcomes the measurement deviation problem caused by insufficient selectivity of electrochemical sensors in environments with multiple coexisting gases, and improves the output accuracy of the sensor.

[0127] By predicting the remaining lifespan of sensors based on the degradation trend of health indicators, a basis for preventive maintenance of sensors can be provided. Furthermore, through the management chain from state perception to accuracy assurance to lifespan prediction, full-cycle health management of sensors can be achieved, improving sensor reliability, reducing the risk of sudden failures, and extending the service life of sensors.

[0128] Example 4 Figure 6 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application. For example... Figure 6 The illustrated electronic device 600 may include: at least one processor 610 ( Figure 6 The diagram shows only one processor, a memory 620, and a computer program 630 stored in the memory 620 and executable on the at least one processor 610, which, when executing the computer program 630, implements the steps in any of the above method embodiments.

[0129] The electronic device 600 may be a robot or other processing device capable of implementing the above methods. This application embodiment does not impose any restrictions on the specific type of electronic device.

[0130] Those skilled in the art will understand that Figure 6 This is merely an example of electronic device 600 and does not constitute a limitation on the electronic device. It may include more or fewer components than shown, or combine certain components, or use different components. For example, electronic device 600 may also include input / output interfaces.

[0131] The processor 610 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASTCs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0132] In some embodiments, the memory 620 may be an internal storage unit, such as a hard disk or RAM. In other embodiments, the memory 620 may be an external storage device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD), or a flash card. Furthermore, the memory 620 may include both internal and external storage units. The memory 620 is used to store the operating system, applications, a boot loader, data, and other programs, such as the program code of the computer program. The memory 620 can also be used to temporarily store data that has been output or will be output.

[0133] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0134] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0135] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0136] This application provides a computer program product that, when run on an electronic device, enables the electronic device to perform the steps described in the various method embodiments above.

[0137] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0138] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0139] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

Claims

1. A sensor status diagnosis method, characterized in that, include: Acquire environmental data and sensor readings of various health indicators; The environmental data and the detection value are input into the health indicator prediction model to obtain the predicted value of the health indicator. The health indicator prediction model adopts a deep autoencoder architecture. The predicted value represents the value of the health indicator when the healthy sensor and the sensor are in the same operating environment. The health status of the sensor is determined based on the difference between the detected value and the predicted value.

2. The method according to claim 1, characterized in that, Determining the health status of the sensor based on the difference between the detected value and the predicted value includes: The reconstruction error is determined based on the difference between the detected value and the predicted value; By comparing the reconstruction error with the current fault threshold, the health status of the sensor is determined based on the degree to which the reconstruction error deviates from the current fault threshold; the health status includes healthy, unhealthy, and faulty.

3. The method according to claim 2, characterized in that, Before determining the health status of the sensor based on the difference between the detected value and the predicted value, the method further includes: The current fault threshold is determined based on the distribution of sample reconstruction errors in the previous training cycle.

4. The method according to claim 3, characterized in that, The current fault threshold is equal to the sum of the mean and multiple standard deviations of the sample reconstruction error.

5. The method according to claim 2, characterized in that, If the sensor is in an unhealthy state, the method further includes: The environmental data and the real-time output value of the sensor are input into the measurement deviation prediction model to obtain the measurement deviation of the sensor; The real-time output value is corrected using the measurement deviation to obtain the initial calibrated output value of the sensor. The output value after the initial correction is corrected using a cross-interference matrix to obtain the corrected output value of the sensor. The cross-interference matrix is ​​used to characterize the effect of cross-interference between sensors.

6. The method according to claim 1, characterized in that, The bottleneck layer of the deep autoencoder includes neurons that correspond one-to-one with the types of health indicators. The process of configuring the activation function on each neuron includes: The activation function of the neuron is configured according to the value range of the health indicator type corresponding to the neuron.

7. The method according to any one of claims 1-6, characterized in that, The method further includes: The detected value is input into the life prediction model to obtain the remaining lifespan of the sensor; If the remaining service life is less than the alarm threshold, an alarm notification will be sent to the user.

8. A sensor status diagnostic device, characterized in that, It includes an acquisition unit, a prediction unit, and a diagnosis unit; The acquisition unit is used to acquire environmental data and the detection values ​​of multidimensional health indicators from sensors; The prediction unit is used to input the environmental data and the detection value into the health indicator prediction model to obtain the predicted value of the health indicator. The health indicator prediction model adopts a deep autoencoder architecture, and the predicted value represents the value of the health indicator when the healthy sensor and the sensor are in the same operating environment. The diagnostic unit is used to determine the health status of the sensor based on the difference between the detected value and the predicted value.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by an electronic device, it implements the method as described in any one of claims 1-7.

Images