Method for monitoring the state of an industrial plant and electronic device

Abstract

This application discloses a condition monitoring method and electronic device for industrial equipment, relating to the field of equipment monitoring technology. The method includes acquiring at least one observation data point from at least one sensor on the industrial equipment, used to characterize the condition of the equipment; extracting features from the at least one observation data point to obtain feature observation vectors corresponding to each sensor; evaluating the reliability weight of each sensor based on each feature observation vector and set evaluation indicators, wherein the set evaluation indicators include the data uncertainty index of each sensor and / or the consistency deviation between different sensors; weighting and fusing each feature observation vector according to the reliability weight of each sensor to obtain a fused observation value; and estimating the condition of the industrial equipment using a nonlinear condition estimation algorithm and the fused observation value to generate a condition estimate. This solution can effectively achieve accurate and reliable estimation of the condition of industrial equipment, providing effective support for health management and predictive maintenance.

Description

Technical Field

[0001] This application relates to the field of equipment monitoring technology, and in particular to a method and electronic device for monitoring the condition of industrial equipment. Background Technology

[0002] In the field of equipment condition monitoring and early warning, existing methods usually assume that all sensors are always reliable or use fixed weights for fusion. These methods cannot cope with the performance degradation of sensors due to aging, drift, electromagnetic interference or local damage, and are prone to misjudging abnormal sensor data as equipment failure, resulting in false alarms. Summary of the Invention

[0003] This application provides a condition monitoring method and electronic device for industrial equipment, which at least solves the problem of false alarms in equipment condition monitoring in related technologies.

[0004] This application provides a method for condition monitoring of industrial equipment, including: Acquire at least one observation data point collected by at least one sensor on an industrial device to characterize the state of the industrial device, and extract features from the at least one observation data point to obtain the feature observation vector corresponding to each sensor; Based on the feature observation vectors and the set evaluation indicators, the credibility weight of each sensor is evaluated, wherein the set evaluation indicators include the data uncertainty index of each sensor and / or the consistency deviation between different sensors; Based on the confidence weight of each sensor, the feature observation vectors are weighted and fused to obtain the fused observation value; The state of the industrial equipment is estimated based on the nonlinear state estimation algorithm and the fused observations, generating a state estimate value.

[0005] This application also provides a condition monitoring device for industrial equipment, comprising: The data processing module is used to acquire at least one observation data collected by at least one sensor on the industrial equipment to characterize the state of the industrial equipment, and to extract features from the at least one observation data to obtain the feature observation vector corresponding to each sensor. A credibility assessment module is used to assess the credibility weight of each sensor based on each of the feature observation vectors and set evaluation indicators, wherein the set evaluation indicators include the data uncertainty index of each sensor and / or the consistency deviation between different sensors. The weighted fusion module is used to perform weighted fusion of the feature observation vectors according to the confidence weight of each sensor to obtain the fused observation value; The state estimation module is used to perform state estimation on the industrial equipment based on the nonlinear state estimation algorithm and the fused observations, and generate state estimation values.

[0006] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described industrial equipment condition monitoring methods.

[0007] This application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of any of the above-described industrial equipment condition monitoring methods.

[0008] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described industrial equipment condition monitoring methods.

[0009] This application achieves several key improvements. First, by extracting features from the observation data, multi-source heterogeneous sensor data is transformed into unified, high-information-density feature observation vectors. This effectively eliminates differences in dimensions, sampling rates, and signal characteristics among different sensors, laying a standardized foundation for subsequent processing. Second, by fusing data uncertainty indicators and inter-sensor consistency deviations, the reliability weights of each sensor are dynamically evaluated. This adaptively identifies and suppresses abnormal data caused by sensor drift, noise, or failure, improving the accuracy of reliability weights under complex operating conditions. Furthermore, by weighting and fusing the feature observation vectors based on their reliability weights, high-reliability sensors are ensured to dominate the fusion result. The resulting fused observations retain the complementary advantages of multi-source information while possessing better anti-interference capabilities and representativeness, providing high-quality input for subsequent state estimation. Finally, a nonlinear state estimation algorithm is used to process the fused observations, more accurately describing the nonlinear characteristics of the dynamic behavior of industrial equipment and overcoming the limitations of traditional linear methods in complex system modeling. The resulting state estimates are not only more accurate but also more sensitive to subtle changes in equipment state, facilitating early anomaly detection and fault warning.

[0010] Therefore, this method constructs a hierarchical, adaptive, and highly robust condition monitoring framework by organically combining feature extraction, dynamic evaluation of confidence weights, weighted fusion, and nonlinear state estimation. It can effectively achieve accurate, reliable, and real-time estimation of the condition of industrial equipment, providing effective support for health management and predictive maintenance. Attached Figure Description

[0011] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a condition monitoring method for industrial equipment provided in this application embodiment; Figure 2 A flowchart illustrating another method for monitoring the condition of industrial equipment provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of a condition monitoring device for industrial equipment provided in an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0014] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0015] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] The specific application environment architecture or specific hardware architecture on which the execution of the condition monitoring method for industrial equipment depends is described here.

[0017] Equipment status monitoring and early warning technology refers to a type of proactive operation and maintenance technology that uses sensors to collect physical quantities during equipment operation in real time, combines signal processing, data analysis and intelligent algorithms to assess the current health status of the equipment, and issues early warning signals before or at an early stage of a failure.

[0018] In the field of equipment condition monitoring and early warning, the relevant technical solutions mainly have the following two limitations: Problem 1: Mismatch between sensor reliability assumptions and actual degradation. Related solutions typically assume all sensors are always in an ideal, reliable state, or use only fixed weights for data fusion, failing to dynamically respond to performance degradation caused by aging, drift, electromagnetic interference, or localized damage. This static fusion mechanism easily misinterprets abnormal sensor output as equipment malfunctions, triggering false alarms and severely impacting the accuracy and reliability of the early warning system.

[0019] Question 2: Fixed noise model exhibits lag response under abrupt changes in state. Conventional Kalman filters typically employ a fixed process noise covariance, which performs well during stable equipment operation. However, when equipment states undergo abrupt changes or are in the early stages of a fault, the filter tends to exhibit slow response and reduced tracking capability due to the mismatch between the system's dynamic characteristics and the preset model. This makes it difficult to capture weak but critical abnormal signals in a timely manner, leading to delayed or missed fault warnings.

[0020] To address the above problems, this application provides a condition monitoring method and electronic device for industrial equipment. The method will now be described in detail, firstly, by referring to the execution flow of the condition monitoring method for industrial equipment.

[0021] Figure 1 This embodiment provides a flowchart of a condition monitoring method for industrial equipment, which may include the following steps.

[0022] S102, acquire at least one observation data collected by at least one sensor on the industrial equipment to characterize the state of the industrial equipment, and extract features from the at least one observation data to obtain the feature observation vector corresponding to each sensor.

[0023] The industrial equipment in this embodiment is an intelligent industrial device with high reliability requirements, requiring condition monitoring and predictive maintenance, such as CNC machine tools, industrial robots, and lithium battery energy storage systems. Multiple heterogeneous sensors are deployed on key components of the industrial equipment (such as spindles, bearing housings, and motor housings). These sensors are of various types, including vibration accelerometers, current transformers, temperature sensors, and acoustic emission probes.

[0024] This embodiment uses at least one sensor deployed on industrial equipment to collect raw physical quantity data reflecting the operating status of the industrial equipment, forming observation data; wherein, each sensor corresponds to its own observation data.

[0025] Heterogeneous sensors have different sampling rates, dimensions, and time bases, resulting in poor data consistency in directly acquired observation data due to inconsistencies in timing, large scale differences, and strong noise interference. Therefore, when extracting features from at least one set of observation data, it is necessary to first synchronize the time of observation data acquired by at least one sensor using hardware clock alignment or software interpolation algorithms. Then, the time-synchronized observation data should undergo at least one of the following processing methods: denoising, normalization, key statistical feature extraction, and time-frequency domain feature extraction, to obtain a feature observation vector with consistent dimensions.

[0026] In practice, hardware clock alignment or software interpolation algorithms can be used to synchronize the observation data from multiple sources, ensuring that all observation data are aligned under a unified time reference and eliminating time deviations between observation data from different sensors.

[0027] For the time-synchronized observation data, a moving average filter and wavelet thresholding method are used for denoising, and then Z-score standardization is used for normalization to eliminate the dimensional influence between different observation data and form observation data with consistent dimensions.

[0028] Based on this, key features are extracted from the observation data. For example, in the time domain, statistical features such as the mean, variance, root mean square, peak value, and kurtosis of the observation data are calculated to characterize the energy and impact properties of the data. For instance, if the observation data is the change in bearing temperature over time, key statistical features are extracted from this data in the time domain to obtain the rate of temperature change per unit time. The rate of temperature change has the following clear physical meaning: it reflects anomalies in thermodynamic processes. An abnormally high rate of temperature change may indicate lubrication failure or increased friction.

[0029] In the frequency domain, the dominant frequency amplitude, centroid frequency, and frequency band energy proportion in the spectrum of the observed data are extracted using Fast Fourier Transform (FFT), revealing the periodicity and harmonic components of the observed data. To further capture non-stationary characteristics, Short-Time Fourier Transform (SFT) or wavelet packet decomposition is introduced to extract the energy concentration and modal complexity characteristics in the joint time-frequency distribution of the observed data.

[0030] After time synchronization, denoising, normalization, key statistical feature extraction, and time-frequency domain feature extraction, the time deviation and dimensional differences between sensors are effectively eliminated, forming a feature vector with a unified time reference, unified dimension, clear physical meaning, and controllable noise. This improves the consistency and representational ability of the feature vector, providing a reliable data foundation for subsequent high-precision fusion and state estimation. Each sensor corresponds to its own feature vector.

[0031] S104. Based on the observation vectors of each feature and the set evaluation indicators, evaluate the credibility weight of each sensor. The set evaluation indicators include the data uncertainty index of each sensor and / or the consistency deviation between different sensors.

[0032] This embodiment can first calculate the current data uncertainty index of each sensor and the consistency deviation of the state estimation results between the current sensor and other sensors within a preset time window based on the feature observation vector of each sensor; then generate the dynamic reliability weight of each sensor based on the above data uncertainty index and the above consistency deviation.

[0033] In a specific embodiment, the confidence weight of each sensor is evaluated based on each feature observation vector and a set evaluation index, which may include: For the current sensor, cluster analysis is performed on multiple feature observation vectors at multiple time points to divide them into at least one data cluster. The proportion of feature observation vectors contained in each data cluster among all feature observation vectors is calculated to determine the probability distribution.

[0034] Specifically, the current sensor is any one of at least one sensors, such as the i-th sensor deployed on an industrial device. Multiple feature observation vectors of the current sensor i at the current time k and multiple sampling times prior to that time are obtained.

[0035] Cluster analysis is performed on multiple feature observation vectors to divide them into several data clusters; in the cluster analysis, the feature observation vectors are used as samples. The proportion of samples contained in each data cluster in the total number of samples is calculated to form a probability distribution. The probability distribution formed by the feature observation vector corresponding to the current sensor i can be expressed as: , where m represents the m-th data cluster.

[0036] Based on the probability distribution, the Shannon information entropy is calculated as the data uncertainty index of the current sensor i (denoted as Hi), referring to the following expression (1).

[0037] (1) The aforementioned uncertainty indicator H i Indicates the first The instantaneous data uncertainty of a sensor indicates that the larger the value, the more chaotic the output data and the lower the reliability.

[0038] Simultaneously, within a sliding time window of a first predetermined length (denoted as L), the feature observation vectors of each sensor are mapped to local state estimates, and the average state estimate of at least one sensor is calculated. The average state estimate of at least one sensor (i.e., the sensor group) is: (2) in, This represents the local state estimate of the i-th sensor at time l. The average state estimate of the sensor group is represented by x, which is an abstract representation of the overall operating state of the production equipment. N represents the total number of sensors.

[0039] Furthermore, based on the local state estimate and the average state estimate, the degree of deviation of each sensor from the behavior of at least one sensor within the sliding time window is calculated to obtain the consistency deviation between different sensors; specifically refer to the following expression (3).

[0040] (3) Among them, D i Let represent the consistency deviation between the i-th sensor and the sensor group within a time window of length L, and k represent the current time.

[0041] Next, based on the set adjustment coefficient, data uncertainty index, and consistency deviation, the confidence score of each sensor is calculated, and the confidence score is normalized to obtain the confidence weight of each sensor.

[0042] Specifically, a preset adjustment coefficient is introduced. and Combining the data uncertainty index Hi and the consistency deviation Di, the confidence score s of the i-th sensor is calculated with reference to the following expression (4). i : (4) Among them, the confidence score s i With data uncertainty index H i Or consistency deviation D i The increase is exponential, which allows for rapid suppression of abnormal sensors.

[0043] The confidence scores of each sensor are normalized to obtain the confidence weight w for each sensor. i Refer to the following expression (5): (5) Among them, w i Indicates the first The credibility weight of each sensor at the current moment represents its overall credibility; s j This represents the confidence score of sensor j other than sensor i.

[0044] It should be noted that by introducing a dual criterion of instantaneous uncertainty based on information entropy and historical consistency deviation based on group consistency, and generating dynamic credibility weights through an exponential decay function, the system can simultaneously perceive instantaneous anomalies (such as sudden noise) and long-term degradation (such as drift or aging) of sensors, thereby achieving rapid suppression of unreliable sensors and precise enhancement of reliable sensors, significantly improving fusion robustness.

[0045] S106. Based on the confidence weight of each sensor, the feature observation vectors are weighted and fused to obtain the fused observation value.

[0046] This embodiment may include: using a linear convex combination method, weighting and summing the feature observation vectors corresponding to at least one sensor in the feature space according to the corresponding confidence weights to obtain fused observation values.

[0047] Specifically, the weighted fusion adopts a linear convex combination method, that is, the feature observation vectors output by each sensor are summed according to their corresponding dynamic confidence weights to ensure that high-confidence sensors contribute more to the fusion results.

[0048] The fusion process is carried out in the feature space, preserving the semantic information of the original observation data and avoiding information loss. At the same time, it ensures that the fused observations have good robustness and representativeness, providing high-quality input for subsequent adaptive filtering.

[0049] This embodiment employs a weighted fusion strategy that satisfies convex combination constraints. While preserving the original semantics of the observation data, it ensures that the fusion result has mathematical stability and physical interpretability. High-reliability sensors contribute more to the fused observations, effectively suppressing interference from low-quality data and improving the data quality of subsequent data processing.

[0050] S108, perform state estimation on industrial equipment based on nonlinear state estimation algorithm and fused observations, and generate state estimate values.

[0051] In one implementation, an adaptive strong tracking filter can be used to implement the nonlinear state estimation algorithm; in a specific example of an adaptive strong tracking filter, a square root occultation Kalman filter can be used. The square root occultation Kalman filter is essentially a nonlinear state estimation algorithm.

[0052] Accordingly, this embodiment includes: fusing observations (represented as z) fused The input is fed into an adaptive strong tracking filter to perform state estimation, obtaining the current state estimate of the industrial equipment (represented as...). The current state estimate is the system state of the industrial equipment at the current sampling time k, predicted by the adaptive strong tracking filter based on the fused observations.

[0053] After estimating the state of industrial equipment based on a nonlinear state estimation algorithm and fused observations to generate state estimates, the method provided in this embodiment further includes: Step S202: The difference between the fused observations and the state estimates is determined as the innovation residual (denoted as V). k ); that is: (6) The innovation residual reflects the degree of matching between the adaptive strong tracking filter and the industrial equipment. Based on this, a scaling factor is defined according to the innovation residual, the preset observation noise covariance matrix, and the preset matrix trace. The scaling factor is used to dynamically adjust the process noise covariance. The initial process noise covariance in the nonlinear state estimation algorithm is updated according to the scaling factor for the next filtering iteration.

[0054] Specifically, refer to the following expression (7), based on the new information residual V k 1. Preset the observation noise covariance matrix R and the preset matrix trace Tr(), and determine the scaling factor. : (7) Based on the scaling factor mentioned above, if the innovation residual energy increases, it indicates a mismatch in the adaptive strong tracking filter, and the process noise covariance should be increased to enhance the tracking capability. If the innovation residual energy is less than the observation noise energy, it indicates that the system is stable and no adjustment is needed; otherwise, the scaling factor should be applied. Noise covariance during amplification.

[0055] Based on the above scaling factors Update the initial process noise covariance Q0 to This is used in the next filtering iteration, enabling the adaptive strong tracking filter to automatically enhance its tracking capability when the state changes abruptly.

[0056] This embodiment uses the feedback of the innovation residual to dynamically scale the process noise covariance, so that the filter maintains a low gain to suppress noise when the industrial equipment is running smoothly, and automatically increases the gain to enhance the tracking capability when the state changes suddenly or an early fault occurs. Thus, it significantly improves the sensitivity and response speed to weak fault signals without sacrificing steady-state accuracy.

[0057] Step S204: Scalarize the new information residual to generate a residual scalar sequence, and calculate the nonlinear entropy index to reflect the sequence complexity based on the residual scalar sequence.

[0058] This embodiment includes: calculating the Euclidean norm of the innovation residual to obtain a residual scalar sequence; calculating the template matching ratio of the residual scalar sequence according to a set embedding dimension within a sliding time window of a second set length; and calculating a nonlinear entropy index to reflect the sequence complexity based on the template matching ratio.

[0059] Specifically, the Euclidean norm of the innovation residuals is calculated to obtain the residual scalar sequence, denoted as r(k), whose expression is: r(k) = |Vk|2(8) In the specific method for calculating the nonlinear entropy index of the residual scalar sequence, within a second predetermined sliding time window (e.g., W), the preset embedding dimension is, for example, 2, and the similarity tolerance is 0.2 times the standard deviation of the current sliding time window. Let the template matching ratios for lengths of 2 and 3 be respectively... and Then the expression for the nonlinear entropy index is (9): (9) The aforementioned nonlinear entropy index, SampEn, stands for Sample Entropy. It is characterized by being nonlinear, nonparametric, and requiring no phase space reconstruction. It is used to measure the regularity or complexity of residual scalar sequences. The larger the value, the more unstable, complex, and unpredictable the residual scalar sequence is. An increase in the value indicates dynamic instability caused by early faults.

[0060] This embodiment constructs a residual scalar sequence based on the norm of the new residual and calculates the nonlinear entropy index. It can effectively capture changes in the dynamic complexity of the system without relying on fault labels or prior models. The nonlinear entropy index is extremely sensitive to the chaos or increased randomness caused by early weak faults, overcoming the shortcomings of traditional linear residual detection methods in terms of insufficient sensitivity under complex working conditions.

[0061] Step S206: Compare the nonlinear entropy index with a preset benchmark threshold under normal operating conditions of the industrial equipment, and trigger an early warning signal when the nonlinear entropy index exceeds the benchmark threshold.

[0062] This embodiment may include: first, acquiring sample entropy values ​​collected during a set period of time under normal operating conditions of industrial equipment; calculating statistical values ​​of sample entropy values ​​within the set period of time; and calculating a baseline threshold for industrial equipment under normal operating conditions based on the statistical values ​​and a set sensitivity coefficient.

[0063] Specifically, under normal operating conditions of industrial equipment, multiple sample entropy values ​​are continuously collected over a set period, and statistical values ​​of these sample entropy values ​​are calculated, such as the mean. with standard deviation Based on this, the baseline threshold for industrial equipment under normal operating conditions is determined as follows: (10) in, >0 represents the set sensitivity coefficient.

[0064] When the nonlinear entropy index exceeds a baseline threshold, the industrial equipment is deemed to be in an abnormal state. That is, when the real-time calculated nonlinear entropy index... When this happens, the industrial equipment is determined to be in an abnormal state.

[0065] The moment when the state is determined to be abnormal is taken as the starting point of the abnormality, and continuous periodic counting is started.

[0066] If the duration of an industrial equipment in an abnormal state exceeds the preset continuous period threshold, such as when the industrial equipment remains in an abnormal state for more than T sampling periods, a level two warning signal will be triggered to recommend shutdown for maintenance.

[0067] If the duration of an abnormal state of industrial equipment does not exceed the continuous period threshold, a Level 1 warning signal is triggered to prompt the user to strengthen monitoring.

[0068] The aforementioned continuous period threshold T is used to distinguish between real faults and transient interference, and can effectively reduce the false alarm rate.

[0069] This embodiment maintains the sample entropy value online and sets an adaptive warning benchmark threshold based on the mean and standard deviation. This enables the warning mechanism to automatically adjust the judgment criteria as the normal operating conditions of industrial equipment slowly drift (such as aging or load changes), effectively avoiding frequent false alarms or missed alarms caused by fixed thresholds under changing operating conditions, and improving the environmental adaptability and reliability of the warning system.

[0070] According to the above embodiments, after triggering a warning signal when the nonlinear entropy index exceeds a set threshold, the method provided in this embodiment may further include: It outputs early warning signals and associates the output with the credibility weight, state estimate and nonlinear entropy index on which it is based, to complete the state monitoring and early warning of industrial equipment.

[0071] Specifically, when an early warning signal is triggered, the following structured information is encapsulated: early warning level, the time of determination of the abnormal state, the set of confidence weights, the state estimate, the current nonlinear entropy index, and the statistical value of the sample entropy. This structured information is then transmitted to the monitoring terminal or cloud platform via the industrial communication protocol MQTT.

[0072] Furthermore, the health status of sensors can be determined based on the distribution of confidence weights, and the faults of industrial equipment can be identified by combining the trends of state transitions and nonlinear entropy indicators, thus achieving intelligent diagnosis.

[0073] In this embodiment, fault diagnosis of sensors and industrial equipment is performed based on confidence weights, state estimates, and nonlinear entropy indices. The implementation method can be referred to the following content.

[0074] Example 1. When multiple sensors are configured on an industrial device, if the confidence weight of the target sensor is lower than that of the other sensors, and the confidence weights of the other sensors are within a first set range, the state estimates are within a second set range, and the nonlinear entropy index is within a third set range, then the target sensor is determined to have a hardware fault, signal drift, or installation abnormality.

[0075] Specifically, Condition 1 states that the confidence weight of the target sensor is lower than that of the other sensors, indicating that the target sensor's weight is significantly low and inconsistent with most other sensors. Condition 2 states that the confidence weights of the other sensors are within the first set range, indicating that the other sensors corroborate each other, their weights are balanced and within a reasonable range, suggesting they are in normal working condition. Condition 3 states that the state estimates of the other sensors are within the second set range, indicating that the state estimates of the other sensors are stable and no abnormal jumps were found, suggesting that the industrial equipment itself is operating stably. Condition 4 states that the nonlinear entropy index of the other sensors is within the third set range, indicating that the nonlinear entropy index has not exceeded the limit, suggesting that the industrial equipment has not experienced early failure or dynamic instability.

[0076] If conditions 2 through 4 are met, it indicates that most sensors are consistent, thus the industrial equipment can be considered normal. Considering that condition 1 is also met, the source of the anomaly can be located to a specific target sensor. Therefore, it can be determined that the target sensor has a hardware fault, signal drift, or improper installation.

[0077] Example 2. If the confidence weights of multiple sensors are all higher than the preset safety weight threshold, and among the state estimates of multiple sensors, there is a value lower than the decrease in the state estimate within a single cycle that exceeds the preset change threshold, and the nonlinear entropy index is continuously higher than the preset first warning entropy threshold, then it is determined that a sudden failure has occurred in the industrial equipment.

[0078] Specifically, Condition 1 states that the credibility weights of multiple sensors are all higher than the preset safety weight threshold, indicating that most sensors are reliable and functioning normally. Based on weight consistency and historical performance, the sensor group has reached a consensus and is considered a reliable data source. Condition 2 states that among the state estimates of multiple sensors, there is a decrease in the state estimate value within a single cycle exceeding a preset change threshold. This indicates a significant deterioration in the core indicator reflecting the health of industrial equipment (i.e., the state estimate value). An abnormal decrease in the state estimate value is the most direct manifestation of equipment failure. Condition 3 states that the nonlinear entropy index of multiple sensors is consistently higher than the first warning entropy threshold. This indicates that the nonlinear entropy index, reflecting the complexity of the industrial equipment's operating state, is continuously abnormal, suggesting that the regularity of the system output is disrupted, and randomness or chaos is enhanced. This is a sensitive indicator of early faults (such as wear and imbalance) causing changes in system dynamics. A nonlinear entropy index consistently higher than the first warning entropy threshold can filter out transient interference.

[0079] When all three conditions are met simultaneously, indicating an anomaly, it can be determined with high confidence that the industrial equipment itself has malfunctioned.

[0080] Example 3. If the nonlinear entropy indices of multiple sensors show an upward trend, and any nonlinear entropy index is lower than the preset second warning entropy threshold, then the industrial equipment is determined to be in a progressive degradation stage, and enhanced monitoring and predictive maintenance are recommended.

[0081] Specifically, Condition 1 indicates that the nonlinear entropy indices of multiple sensors show an upward trend, signifying a decrease in the regularity and an increase in the complexity of the industrial equipment's operating status. This is typically a direct reflection of slow physical processes such as component wear, increased clearance, decreased lubrication performance, and material fatigue. Condition 2 indicates that any nonlinear entropy index is below the second warning entropy threshold, which serves as the boundary distinguishing between normal fluctuations, gradual degradation, and failure. Condition 2 suggests that the industrial equipment has not yet reached the level to trigger an emergency alarm and is still in a sub-healthy or early-stage decline state, representing the optimal window for preventative intervention.

[0082] If the above two conditions are met, the industrial equipment is determined to be in a progressive degradation stage, and enhanced monitoring and predictive maintenance are recommended.

[0083] The above example, by integrating three pieces of evidence—confidence weights, state estimates and their evolution trends, and nonlinear entropy indices—not only provides early warning results but also supports the differentiation of root causes of failures, such as sensor failures or industrial equipment failures. This can effectively improve diagnostic efficiency, decision-making scientificity, and overall system interpretability.

[0084] In the above embodiments, by providing structured outputs containing multi-dimensional diagnostic criteria such as credibility weights, state estimates, and nonlinear entropy indices, not only are early warning results provided, but root cause analysis of faults is also supported: maintenance personnel can use this to distinguish between sensor failures and equipment malfunctions, greatly shortening diagnostic time, improving the efficiency and scientific nature of intelligent maintenance and decision-making, and enhancing the overall interpretability of the system.

[0085] In summary, the industrial equipment condition monitoring method provided in this disclosure includes: acquiring at least one observation data point collected by at least one sensor on the industrial equipment to characterize the condition of the industrial equipment, and extracting features from the at least one observation data point to obtain feature observation vectors corresponding to each sensor; evaluating the reliability weight of each sensor based on each feature observation vector and a set evaluation index, wherein the set evaluation index includes the data uncertainty index of each sensor and / or the consistency deviation between different sensors; performing weighted fusion of each feature observation vector according to the reliability weight of each sensor to obtain a fused observation value; and performing condition estimation of the industrial equipment according to a nonlinear condition estimation algorithm and the fused observation value to generate a condition estimate value.

[0086] Based on this technical solution, firstly, feature extraction is performed on the observation data to transform multi-source heterogeneous sensor data into unified, high-information-density feature observation vectors. This effectively eliminates differences in dimensions, sampling rates, and signal characteristics among different sensors, laying a standardized foundation for subsequent processing. Secondly, by fusing data uncertainty indicators and inter-sensor consistency deviations, the reliability weight of each sensor is dynamically evaluated. This adaptively identifies and suppresses abnormal data caused by sensor drift, noise, or failure, improving the accuracy of reliability weights under complex operating conditions. Furthermore, the feature observation vectors are weighted and fused according to their reliability weights, ensuring that high-reliability sensors dominate the fusion result. The resulting fused observations retain the complementary advantages of multi-source information while possessing better anti-interference capabilities and representativeness, providing high-quality input for subsequent state estimation. Finally, a nonlinear state estimation algorithm is used to process the fused observations, which can more accurately describe the nonlinear characteristics in the dynamic behavior of industrial equipment, overcoming the limitations of traditional linear methods in complex system modeling. The resulting state estimates are not only more accurate but also more sensitive to subtle changes in equipment state, facilitating early anomaly detection and fault warning.

[0087] Therefore, this method constructs a hierarchical, adaptive, and highly robust condition monitoring framework by organically combining feature extraction, dynamic evaluation of confidence weights, weighted fusion, and nonlinear state estimation. It can effectively achieve accurate, reliable, and real-time estimation of the condition of industrial equipment, providing effective support for health management and predictive maintenance.

[0088] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0089] Embodiments of this application also provide a condition monitoring device for industrial equipment, comprising: The data processing module 310 is used to acquire at least one observation data collected by at least one sensor on the industrial equipment to characterize the state of the industrial equipment, and to extract features from the at least one observation data to obtain the feature observation vector corresponding to each sensor. The credibility assessment module 320 is used to assess the credibility weight of each of the sensors based on the feature observation vectors and the set evaluation indicators, wherein the set evaluation indicators include the data uncertainty index of each sensor and / or the consistency deviation between different sensors. The weighted fusion module 330 is used to perform weighted fusion of the feature observation vectors according to the confidence weight of each sensor to obtain the fused observation value; The state estimation module 340 is used to perform state estimation on the industrial equipment based on the nonlinear state estimation algorithm and the fused observations, and generate state estimation values.

[0090] In one embodiment, the credibility assessment module 320 is further configured to: For the current sensor, cluster analysis is performed on multiple feature observation vectors at multiple times to divide them into at least one data cluster; wherein, the current sensor is any one of the at least one sensors; The probability distribution is determined by calculating the proportion of the feature observation vector contained in each data cluster among all the feature observation vectors. Based on the probability distribution, Shannon information entropy is calculated as an indicator of the data uncertainty of the current sensor. Within a first predetermined sliding time window, the feature observation vectors of each sensor are mapped to local state estimates, and the average state estimate of at least one sensor is calculated. Based on the local state estimate and the average state estimate, the degree of deviation of the behavior of each sensor from that of the at least one sensor within the sliding time window is calculated to obtain the consistency deviation between different sensors; Based on the set adjustment coefficient, the data uncertainty index, and the consistency deviation, the confidence score of each sensor is calculated, and the confidence score is normalized to obtain the confidence weight of each sensor.

[0091] In one embodiment, the device further includes: The innovation residual determination module is used to determine the difference between the fused observation and the state estimate as the innovation residual; The entropy index determination module is used to scalarize the innovation residual, generate a residual scalar sequence, and calculate a nonlinear entropy index to reflect the sequence complexity based on the residual scalar sequence. The early warning triggering module is used to compare the nonlinear entropy index with a preset benchmark threshold of the industrial equipment under normal operating conditions, and to trigger an early warning signal when the nonlinear entropy index is greater than the benchmark threshold.

[0092] In one embodiment, the warning triggering module is further configured to: Obtain the sample entropy value collected during a set time period under normal operating conditions of the industrial equipment; Calculate the statistical value of the sample entropy value within the specified time period; The baseline threshold of the industrial equipment under normal operating conditions is calculated based on the statistical values ​​and the set sensitivity coefficient. When the nonlinear entropy index is greater than the benchmark threshold, the industrial equipment is determined to be in an abnormal state. The time when the state is deemed abnormal is taken as the starting point of the abnormality, and continuous periodic counting is started. If the duration of the abnormal state of the industrial equipment exceeds the preset continuous period threshold, a level two warning signal is triggered to recommend shutdown for maintenance. If the duration of the abnormal state of the industrial equipment does not exceed the continuous period threshold, a first-level early warning signal is triggered to prompt the user to strengthen monitoring.

[0093] In one embodiment, the device further includes a fault diagnosis module, which is used for: When multiple sensors are configured on the industrial equipment, if the confidence weight of the target sensor is lower than the confidence weight of the other sensors, and the confidence weight of the other sensors is within a first set range, the state estimate is within a second set range, and the nonlinear entropy index is within a third set range, then it is determined that the target sensor has a hardware fault, signal drift, or installation abnormality. If the confidence weights of the multiple sensors are all higher than the preset safety weight threshold, and among the state estimates of the multiple sensors, there is a value lower than the decrease value of the state estimate within a single cycle that exceeds the preset change threshold, and the nonlinear entropy index is continuously higher than the preset first warning entropy threshold, then it is determined that the industrial equipment has experienced a sudden failure. If the nonlinear entropy indices of the multiple sensors show an upward trend, and any nonlinear entropy index is lower than the preset second warning entropy threshold, then the industrial equipment is determined to be in a progressive degradation stage, and a suggestion is made to strengthen monitoring and predictive maintenance.

[0094] In one embodiment, the entropy index determination module is further configured to: Calculate the Euclidean norm of the innovation residuals to obtain the residual scalar sequence; Within a second set sliding time window, the template matching ratio of the residual scalar sequence is calculated according to a set embedding dimension; The nonlinear entropy index used to reflect sequence complexity is calculated based on the template matching ratio.

[0095] In one embodiment, the apparatus further includes a parameter update module, which is used to: A scaling factor is defined based on the new information residual, the preset observation noise covariance matrix, and the preset matrix trace, wherein the scaling factor is used to adjust the process noise covariance; The initial process noise covariance in the nonlinear state estimation algorithm is updated according to the scaling factor for the next filtering iteration.

[0096] In one embodiment, the data processing module 310 is further configured to: Time synchronization of observation data acquired by at least one sensor is achieved through hardware clock alignment or software interpolation algorithms. Perform at least one of the following processing steps on at least one of the time-synchronized observation data: denoising, normalization, key statistical feature extraction, and time-frequency domain feature extraction, to obtain a feature observation vector with consistent dimensions.

[0097] In one embodiment, the weighted fusion module 330 is further configured to: By employing a linear convex combination method, the feature observation vectors corresponding to at least one sensor are weighted and summed in the feature space according to their corresponding confidence weights to obtain fused observation values.

[0098] For a description of the features in the embodiment corresponding to the condition monitoring device for industrial equipment, please refer to the relevant description of the embodiment corresponding to the condition monitoring method for industrial equipment, which will not be repeated here.

[0099] Embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above embodiments of the condition monitoring method for industrial equipment.

[0100] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above embodiments of the condition monitoring method for industrial equipment.

[0101] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0102] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described embodiments of the industrial equipment condition monitoring method.

[0103] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described embodiments of the industrial equipment condition monitoring method.

[0104] Any of the components, modules, units, parts, methods, and operations described herein can be implemented using software, firmware, hardware (e.g., fixed logic circuitry), manual processing, or any combination thereof. Alternatively or additionally, any functionality described herein can be executed at least in part by one or more hardware logic components, such as, but not limited to, a central processing unit (CPU), a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), an application-specific standard product (ASSP), a system-on-a-chip (SoC), a complex programmable logic device (CPLD), a microprocessor (MCU), etc. The terms "system," "computing device," or "apparatus" as used herein encompass various means, devices, and machines for processing data, including, for example, one or more programmable processors, computers, SoCs, or combinations thereof. The apparatus may also include code that creates an execution environment for the computer program in question, such as code constituting processor firmware, a protocol stack, a database management system, an operating system, a cross-platform runtime environment, a virtual machine, or one or more combinations thereof. The aforementioned computer program (also known as a program, software, software application, app, script, or code) can be written in any form of programming language, including compiled or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, object, or other unit suitable for a computing environment.

[0105] Those skilled in the art will further 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, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. 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.

[0106] The above provides a detailed description of the condition monitoring method and electronic device for industrial equipment provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for condition monitoring of industrial equipment, characterized in that, include: Acquire at least one observation data point collected by at least one sensor on an industrial device to characterize the state of the industrial device, and extract features from the at least one observation data point to obtain the feature observation vector corresponding to each sensor; Based on the feature observation vectors and the set evaluation indicators, the credibility weight of each sensor is evaluated, wherein the set evaluation indicators include the data uncertainty index of each sensor and / or the consistency deviation between different sensors; Based on the confidence weight of each sensor, the feature observation vectors are weighted and fused to obtain the fused observation value; The state of the industrial equipment is estimated based on the nonlinear state estimation algorithm and the fused observations, generating a state estimate value. The difference between the fused observation and the state estimate is defined as the innovation residual. The new information residual is scalarized to generate a residual scalar sequence, and a nonlinear entropy index reflecting the sequence complexity is calculated based on the residual scalar sequence. The nonlinear entropy index is compared with a preset benchmark threshold under normal operating conditions of the industrial equipment, and an early warning signal is triggered when the nonlinear entropy index is greater than the benchmark threshold. The step of evaluating the credibility weight of each sensor based on each of the feature observation vectors and the set evaluation index includes: For the current sensor, cluster analysis is performed on multiple feature observation vectors at multiple times to divide them into at least one data cluster; wherein, the current sensor is any one of the at least one sensors; The probability distribution is determined by calculating the proportion of the feature observation vector contained in each data cluster among all the feature observation vectors. Based on the probability distribution, Shannon information entropy is calculated as an indicator of the data uncertainty of the current sensor. Within a first predetermined sliding time window, the feature observation vectors of each sensor are mapped to local state estimates, and the average state estimate of at least one sensor is calculated. Based on the local state estimate and the average state estimate, the degree of deviation of the behavior of each sensor from that of the at least one sensor within the sliding time window is calculated to obtain the consistency deviation between different sensors; Based on the set adjustment coefficient, the data uncertainty index, and the consistency deviation, the confidence score of each sensor is calculated, and the confidence score is normalized to obtain the confidence weight of each sensor.

2. The condition monitoring method for industrial equipment according to claim 1, characterized in that, The step of comparing the nonlinear entropy index with a preset benchmark threshold under normal operating conditions of the industrial equipment, and triggering a warning signal when the nonlinear entropy index exceeds the benchmark threshold, includes: Obtain the sample entropy value collected during a set time period under normal operating conditions of the industrial equipment; Calculate the statistical value of the sample entropy value within the specified time period; The baseline threshold of the industrial equipment under normal operating conditions is calculated based on the statistical values ​​and the set sensitivity coefficient. When the nonlinear entropy index is greater than the benchmark threshold, the industrial equipment is determined to be in an abnormal state. The time when the state is deemed abnormal is taken as the starting point of the abnormality, and continuous periodic counting is started. If the duration of the abnormal state of the industrial equipment exceeds the preset continuous period threshold, a level two warning signal is triggered to recommend shutdown for maintenance. If the duration of the abnormal state of the industrial equipment does not exceed the continuous period threshold, a first-level early warning signal is triggered to prompt the user to strengthen monitoring.

3. The condition monitoring method for industrial equipment according to claim 1, characterized in that, After triggering the warning signal when the nonlinear entropy index is greater than the benchmark threshold, the method further includes: When multiple sensors are configured on the industrial equipment, if the confidence weight of the target sensor is lower than the confidence weight of the other sensors, and the confidence weight of the other sensors is within a first set range, the state estimate is within a second set range, and the nonlinear entropy index is within a third set range, then it is determined that the target sensor has a hardware fault, signal drift, or installation abnormality. If the confidence weights of the multiple sensors are all higher than the preset safety weight threshold, and among the state estimates of the multiple sensors, there is a value lower than the decrease value of the state estimate within a single cycle that exceeds the preset change threshold, and the nonlinear entropy index is continuously higher than the preset first warning entropy threshold, then it is determined that the industrial equipment has experienced a sudden failure. If the nonlinear entropy indices of the multiple sensors show an upward trend, and any nonlinear entropy index is lower than the preset second warning entropy threshold, then the industrial equipment is determined to be in a progressive degradation stage, and a suggestion is made to strengthen monitoring and predictive maintenance.

4. The condition monitoring method for industrial equipment according to claim 1, characterized in that, The step of scalarizing the innovation residual to generate a residual scalar sequence, and calculating a nonlinear entropy index reflecting the sequence complexity based on the residual scalar sequence, includes: Calculate the Euclidean norm of the innovation residuals to obtain the residual scalar sequence; Within a second set sliding time window, the template matching ratio of the residual scalar sequence is calculated according to a set embedding dimension; The nonlinear entropy index used to reflect sequence complexity is calculated based on the template matching ratio.

5. The condition monitoring method for industrial equipment according to claim 1, characterized in that, After determining the difference between the fused observation and the state estimate as the innovation residual, the method further includes: A scaling factor is defined based on the new information residual, the preset observation noise covariance matrix, and the preset matrix trace, wherein the scaling factor is used to adjust the process noise covariance; The initial process noise covariance in the nonlinear state estimation algorithm is updated according to the scaling factor for the next filtering iteration.

6. The condition monitoring method for industrial equipment according to claim 1, characterized in that, The step of extracting features from the at least one observation data to obtain the feature observation vector corresponding to each of the sensors includes: Time synchronization of observation data acquired by at least one sensor is achieved through hardware clock alignment or software interpolation algorithms. Perform at least one of the following processing steps on at least one of the time-synchronized observation data: denoising, normalization, key statistical feature extraction, and time-frequency domain feature extraction, to obtain a feature observation vector with consistent dimensions.

7. The condition monitoring method for industrial equipment according to claim 1, characterized in that, The step of weighting and fusing the feature observation vectors according to the confidence weights of each sensor to obtain the fused observation value includes: By employing a linear convex combination method, the feature observation vectors corresponding to at least one sensor are weighted and summed in the feature space according to their corresponding confidence weights to obtain fused observation values.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the condition monitoring method for industrial equipment as described in any one of claims 1 to 7.

Images