Method, system and equipment for determining running state of equipment

Abstract

The invention belongs to the field of equipment operation and maintenance, particularly relates to a method, a system and equipment for determining an equipment operation state, and aims to solve the problems of lagging fault discovery, dependence on artificial experience, incomplete inspection, high subjectivity and incapability of realizing early warning and predictive maintenance in the prior art. The method comprises the steps that acoustic sensor signals are collected and preprocessed; extracting an envelope spectrum, spectrum energy and modulation characteristics; inputting the features into a trained machine learning model to obtain an abnormal score; dynamically updating a threshold capable of adaptively reflecting the statistical distribution characteristics of the equipment state based on the latest and historical score data; and judging the state of the equipment by comparing the score with a threshold value, and generating an alarm and maintenance suggestion when the equipment is abnormal. According to the invention, intelligent monitoring and self-adaptive diagnosis of the equipment state are realized, and early warning and predictive maintenance are realized.

Description

Technical Field

[0001] The present invention belongs to the field of equipment operation and maintenance, and particularly relates to a method, system and equipment for determining the operating status of equipment. Background Art

[0002] Flight simulators, particularly the highest-level full flight simulators (such as FFS Level D), are essential components of modern pilot training. They utilize a complex six-degree-of-freedom (6-DOF) motion system to provide pilots with a highly realistic sense of aerial movement, making them a crucial platform for high-risk training, emergency procedure drills, and aircraft conversion training. The core actuator of this motion system is typically a parallel mechanism consisting of six high-performance electro-hydraulic or electric servo actuators. These actuators, under computer control, perform precise coordinated telescopic motion to drive the simulated cockpit platform for a combination of translational and rotational motion in three dimensions.

[0003] These actuators are complex in structure, subject to demanding operating conditions, and subject to highly variable loads. Their health directly impacts the reliability, safety, and training effectiveness of simulator operations. Currently, the industry generally employs a combination of daily manual inspections and scheduled maintenance checks to monitor and maintain the operating status of these critical components. Manual inspections primarily rely on the maintenance engineer's sensory experience (such as listening for unusual noises and observing for signs of leaks) and simple instrument measurements; scheduled maintenance requires more in-depth inspections during planned downtimes.

[0004] However, this traditional model has many inherent flaws: First, it has significant hysteresis. It can only be detected when a fault develops to a certain level (such as producing noticeable abnormal noise, performance degradation, or leakage). This makes it impossible to effectively capture early signs of progressive deterioration, such as micro-pitting of bearings, initial wear of seals, and slight valve core sticking. This leads to increased maintenance costs and a higher risk of unplanned downtime.

[0005] Secondly, its coverage and continuity are insufficient. Manual inspections are discrete and have a low sampling rate. It is difficult to cover all states of the actuator under all operating conditions and long-term operation. It is very easy to miss intermittent faults or transient abnormal events that only occur under specific operating conditions.

[0006] Third, it is highly subjective and lacks quantitative standards. Determining whether something is abnormal relies heavily on the maintenance personnel's personal experience and hearing. There is a lack of consistent, objective criteria across personnel, making it difficult to effectively transfer experience and resulting in low accuracy in diagnosing complex faults.

[0007] Fourth, the inspection depth is limited. Relying on sensory perception and simple tools, it is impossible to effectively detect the condition of the actuator's internal components (such as gears and bearing contact surfaces), and it is powerless to detect specific acoustic characteristic changes caused by early internal wear.

[0008] Finally, this model is inefficient and disruptive, requiring dedicated manpower and valuable training time for downtime maintenance. It is difficult to fundamentally achieve the transition from preventive maintenance and post-repair to predictive maintenance, and it is impossible to optimize maintenance plans based on the actual health status of components.

[0009] Based on this, the present invention proposes a method, system and device for determining the operating status of a device. Summary of the Invention

[0010] In order to solve the above-mentioned problems in the prior art, namely, the problems that the prior art has delayed fault detection, reliance on manual experience, incomplete inspection, strong subjectivity and inability to achieve early warning and predictive maintenance, the present invention provides a method, system and device for determining the operating status of equipment.

[0011] A first aspect of the present invention provides a method for determining an operating status of a device, comprising: Collecting raw acoustic signals from acoustic sensors; Preprocessing the original acoustic signal, including filtering and noise reduction operations, to extract the target frequency band signal and suppress environmental noise; Extracting a set of predefined acoustic features from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features; Inputting the acoustic features into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions; Dynamically update the threshold for determining abnormal status based on the latest acquired comprehensive abnormality score sequence and historical abnormality score data. The threshold can adaptively reflect the statistical distribution characteristics of the equipment operation status. Comparing the currently obtained abnormality score with the threshold, and determining whether the device is in an abnormal state based on the comparison result; If it is determined to be abnormal, an alarm signal is generated and maintenance suggestions are output.

[0012] Furthermore, the original acoustic signal from the acoustic sensor is collected by: A vibration sensor is installed on the surface of the device housing, a reference microphone is deployed in the acoustic environment of the device, and signals from the vibration sensor and the reference microphone are synchronously collected as original acoustic signals.

[0013] Furthermore, a set of predefined acoustic features is extracted from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features, and the method is as follows: Performing envelope analysis on the preprocessed signal to obtain its envelope spectrum; Calculating the maximum value of the amplitude within a predetermined frequency range of the envelope spectrum to characterize the intensity of a specific mechanical shock; wherein the specific mechanical shock at least includes a gear meshing shock; Calculating the ratio of the energy of the envelope spectrum within a preset sideband frequency range to the total energy of the envelope spectrum within the full frequency band to indicate the significance of the modulation effect; Calculating the ratio of the energy of the original spectrum in a preset high frequency band to the total energy of the original spectrum in the full frequency band to represent the energy proportion of the high frequency friction or impact noise; The calculated maximum values, sideband energy ratios, and high frequency energy ratios are combined into a set of predefined acoustic eigenvectors.

[0014] Furthermore, the acoustic features are input into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions, and the method is as follows: After the equipment is started up and is confirmed to be in a trouble-free operating state, the acoustic characteristic data is continuously collected during a predetermined learning phase; When the amount of the collected fault-free acoustic feature data reaches a preset minimum quantity requirement, the data is output as a data set, and the data set is used to train a support vector machine model and a Gaussian mixture model; The acoustic features extracted in the real-time operation are respectively input into the support vector machine model and the Gaussian mixture model to calculate the corresponding first anomaly score and second anomaly score; A weighted fusion calculation is performed on the first anomaly score and the second anomaly score to obtain a comprehensive anomaly score.

[0015] Furthermore, based on the latest acquired anomaly score sequence and historical anomaly score data, the threshold for judging the abnormal state is dynamically updated as follows: Continuously obtain the comprehensive anomaly scores generated by real-time calculations to form the latest anomaly score sequence; Merging the latest anomaly score sequence with historically stored normal operating condition anomaly score data to generate an updated anomaly score dataset; Calculating the statistical mean and statistical standard deviation of the anomaly scores based on the updated anomaly score dataset; Based on the calculated statistical mean and statistical standard deviation, combined with the current running time and load status, the threshold generation calculation is performed again to update the threshold.

[0016] Furthermore, based on the calculated statistical mean and statistical standard deviation, combined with the current running time and load status, the threshold generation calculation is performed again. The method is: Obtain the calculated updated statistical mean and updated statistical standard deviation; Obtaining a load sensitivity factor corresponding to the accumulated running time of the current device, wherein the value of the load sensitivity factor decreases according to a predetermined rule as the running time increases; Obtaining a load deviation coefficient representing a difference between a current operating load state and a reference load state; The updated statistical mean, the updated statistical standard deviation of a preset multiple, the product of the load sensitivity factor and the load deviation coefficient are added together to generate an updated adaptive threshold.

[0017] Furthermore, the currently obtained abnormality score is compared with the threshold, and whether the device is in an abnormal state is determined based on the comparison result, and the method is as follows: Compare the comprehensive anomaly score calculated in real time with the current threshold; If the comprehensive abnormality score is greater than the threshold, generating a first determination signal indicating that the health status is abnormal; If the comprehensive abnormality score is less than or equal to the threshold, generating a second determination signal indicating that the health status is normal; Based on the generated determination signal, the final determination result of the real-time health status of the device is output.

[0018] Furthermore, if it is determined to be abnormal, an alarm signal is generated and maintenance suggestions are output. The method is as follows: If the final determination result is that the health status is abnormal, an alarm signal with a predetermined level is generated; Based on the numerical value of the comprehensive abnormality score and the degree of degradation of a specific indicator in the acoustic feature, selecting and generating a corresponding maintenance suggestion from a predetermined set of maintenance measures; Output the alarm signal and the maintenance suggestion.

[0019] Another aspect of the present invention provides a system for determining the operating status of a device. Based on a method for determining the operating status of a device, the system includes: a signal acquisition module configured to collect raw acoustic signals from the acoustic sensor; a preprocessing module configured to preprocess the original acoustic signal, including filtering and noise reduction operations, to extract the target frequency band signal and suppress environmental noise; a feature extraction module configured to extract a set of predefined acoustic features from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features; an anomaly score calculation module configured to input the acoustic features into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions; A threshold update module is configured to dynamically update the threshold for determining abnormal status based on the latest acquired comprehensive abnormality score sequence and historical abnormality score data, wherein the threshold can adaptively reflect the statistical distribution characteristics of the device operating status; a state judgment module configured to compare the currently obtained abnormality score with the threshold value and judge whether the device is in an abnormal state according to the comparison result; The maintenance module is configured to generate an alarm signal and output maintenance suggestions if it is determined to be abnormal.

[0020] A third aspect of the present invention provides an electronic device, comprising: at least one processor; and a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned method for determining the operating status of a device.

[0021] Beneficial effects of the present invention: By employing dual-channel synchronous signal acquisition and adaptive noise reduction processing, the present invention effectively separates and enhances the acoustic characteristics of actuator operation, providing high-quality input signals for subsequent analysis. By extracting features such as meshing impact intensity, sideband energy ratio, and high-frequency energy proportion from the envelope spectrum, the present invention achieves targeted capture and quantitative characterization of internal fault characteristics such as gear wear. A normal acoustic baseline is established through a self-learning architecture that integrates a first-class support vector machine and a Gaussian mixture model, enabling the system to adapt to fluctuations in normal operating conditions. By combining real-time anomaly scoring, historical statistics, and load status to dynamically update judgment thresholds, the monitoring system can adapt to different equipment operating stages and load conditions, maintaining high detection sensitivity and low false alarm rates. Ultimately, through sliding window decision-making and a multi-level alarm mechanism, reliable judgment and timely warning of abnormal conditions were achieved. The entire technical solution formed a complete closed loop from signal perception, feature extraction, intelligent analysis, to decision output, ultimately achieving real-time, accurate, and reliable monitoring of the actuator's health status. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 It is a flow chart of a method for determining the operating status of a device according to the present invention. DETAILED DESCRIPTION

[0023] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the relevant invention are shown in the accompanying drawings.

[0024] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0025] A first embodiment of the present invention provides a method for determining an operating status of a device, comprising: Step S10, collecting original acoustic signals from the acoustic sensor; Step S20, preprocessing the original acoustic signal, including filtering and noise reduction operations, to extract the target frequency band signal and suppress environmental noise; Step S30, extracting a set of predefined acoustic features from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features; Step S40: inputting the acoustic features into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions; Step S50: dynamically updating the threshold for determining abnormal status based on the latest acquired comprehensive abnormality score sequence and historical abnormality score data, wherein the threshold can adaptively reflect the statistical distribution characteristics of the device operating status; Step S60, comparing the currently obtained abnormality score with the threshold, and determining whether the device is in an abnormal state based on the comparison result; If it is determined to be abnormal, an alarm signal is generated and maintenance suggestions are output.

[0026] In order to more clearly illustrate a method for determining the operating status of a device according to the present invention, the following Figure 1 Each step in the embodiment of the present invention is described in detail as follows: Step S10, collecting original acoustic signals from the acoustic sensor; In this embodiment, a vibration sensor is installed on the surface of the device housing, and a reference microphone is deployed in the acoustic environment of the device. Signals from the vibration sensor and the reference microphone are synchronously collected as original acoustic signals.

[0027] The device is an electric servo actuator.

[0028] Specifically, a vibration sensor mounted on the outer shell of the electric servo actuator and a reference microphone deployed in the device's acoustic environment simultaneously collect raw acoustic signals from both sensors. The vibration sensor directly captures the acoustic signal generated by the device's housing, while the reference microphone collects background noise signals from the environment. The two are strictly synchronized in time, providing the raw data foundation for subsequent signal processing and analysis.

[0029] In this embodiment, the vibration sensor is an ICP type accelerometer installed on the actuator housing with a sampling rate of ≥40kHz. The reference microphone is an omnidirectional microphone deployed in the area simulating the cabin noise source; the synchronization accuracy of the two sensor signals is ≤10μs.

[0030] Step S20, preprocessing the original acoustic signal, including filtering and noise reduction operations, to extract the target frequency band signal and suppress environmental noise; In this embodiment, the collected original acoustic signal is first preprocessed, which mainly includes two steps: filtering and noise reduction. A 6th-order Butterworth filter with a passband frequency range of [5kHz, 15kHz] is used to bandpass filter the original signal. Its transfer function is: ; Where s is a complex frequency variable, The center frequency is used to extract the high-frequency target frequency band signal related to the engagement and wear of the actuator cylinder gear.

[0031] In order to suppress the interference of environmental noise, an adaptive noise cancellation technology based on the least mean square algorithm is further adopted. The algorithm uses the environmental noise collected by the reference microphone as the reference input. , taking the filtered signal of the vibration sensor as the expected response , by iteratively updating the filter weight vector To minimize the error signal The weight update and error calculation follow the following formula: ; ; in, μ is the step size factor, and its value range is [0.01, 0.1]. This process can effectively offset the environmental noise component and finally output the preprocessed pure target acoustic signal.

[0032] Step S30, extracting a set of predefined acoustic features from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features; The specific steps include: Step S31, performing envelope analysis on the preprocessed signal to obtain its envelope spectrum; Step S32, calculating the maximum value of the amplitude within a predetermined frequency range of the envelope spectrum to characterize the intensity of a specific mechanical shock; wherein the specific mechanical shock at least includes a gear meshing shock; Step S33, calculating the ratio of the energy of the envelope spectrum within the preset sideband frequency range to the total energy of the envelope spectrum within the full frequency band, to indicate the significance of the modulation effect; Step S34, calculating the ratio of the energy of the original spectrum in a preset high frequency band to the total energy of the original spectrum in the full frequency band, to represent the energy proportion of the high frequency friction or impact noise; In step S35 , the calculated maximum values, sideband energy ratios, and high-frequency energy ratios are combined into a set of predefined acoustic eigenvectors.

[0033] In this embodiment, the preprocessed signal is first subjected to envelope analysis to obtain its envelope spectrum. This process extracts the analytical signal of the signal through Hilbert transform and calculates its modulus value, thereby obtaining an envelope line that reflects the law of signal amplitude change. The envelope line is then subjected to Fourier transform to obtain the final envelope spectrum.

[0034] On the calculated envelope spectrum, the maximum amplitude value is searched and recorded within the range near the preset meshing frequency. This maximum value is defined as the meshing impact intensity, which is used to quantify the intensity level of a specific impact event generated in the mechanical system.

[0035] Furthermore, the ratio of the envelope spectrum's energy within the preset sideband frequency range of 1kHz to 3kHz to its total energy across the full frequency band is calculated. This ratio, defined as the sideband energy ratio, directly reflects the significance of the amplitude modulation effect caused by the fault. Simultaneously, the ratio of the original spectrum's energy within the high-frequency band of 5kHz to 15kHz to its total energy across the full frequency band is calculated. This ratio, defined as the high-frequency energy fraction, characterizes the energy proportion of the high-frequency noise component generated by abnormal friction or impact in the entire acoustic signal.

[0036] Finally, the three quantitative indicators of meshing impact strength, sideband energy ratio, and high-frequency energy proportion are combined into a multidimensional vector, which constitutes a set of predefined acoustic feature vectors for subsequent state identification and fault diagnosis.

[0037] Step S40: inputting the acoustic features into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions; The specific steps include: Step S41, after the device is started and confirmed to be in a fault-free operating state, during a predetermined learning phase, continuously collecting acoustic feature data; Step S42: When the amount of the collected fault-free acoustic feature data reaches a preset minimum amount requirement, the data is output as a data set, and the data set is used to train a support vector machine model and a Gaussian mixture model; Step S43: inputting the acoustic features extracted in real-time operation into the support vector machine model and the Gaussian mixture model respectively to calculate the corresponding first anomaly score and second anomaly score; Step S44: Perform a weighted fusion calculation on the first anomaly score and the second anomaly score to obtain a comprehensive anomaly score.

[0038] During implementation, the specific process is as follows: After the equipment is started and confirmed to be in the initial learning phase of stable and trouble-free operation, the system begins to continuously collect acoustic feature data. This predetermined learning phase is usually set to the first 50 hours of operation after the equipment is put into operation. When the amount of accumulated trouble-free feature data reaches the preset minimum sample quantity requirement, that is, 500 sets of feature vectors, the data set is output and used to train two benchmark models in parallel: a first-class support vector machine model and a Gaussian mixture model. The first-class support vector machine model uses a radial basis function as the kernel function, and its kernel function parameter γ is set to 0.1, and the outlier ratio parameter ν is set to 0.01. The Gaussian mixture model sets its number of components to 3 to better model the probability distribution of normal operating data.

[0039] After training is complete, the entire fault-free dataset is re-input into the two trained models, and the anomaly scores are calculated. The first anomaly score output by the first-class support vector machine model ranges from -1 to 1, with larger values ​​indicating greater anomaly. The second anomaly score output by the Gaussian mixture model is the negative log-likelihood value, with larger values ​​also indicating greater anomaly. Based on these two anomaly scores for all normal samples, their statistical mean and standard deviation are calculated.

[0040] The system then dynamically calculates and updates the adaptive threshold for health status assessment based on the calculated statistical mean and standard deviation, a preset load sensitivity factor that decays exponentially based on the device's operating time, and the deviation coefficient between the current real-time load and the baseline load state. This threshold is calculated by adding three times the standard deviation to the mean, multiplying the load sensitivity factor by the load deviation coefficient.

[0041] During real-time monitoring, the system simultaneously feeds each newly extracted acoustic feature vector into a deployed first-class support vector machine model and a Gaussian mixture model to calculate the corresponding first and second anomaly scores. Finally, a weighted geometric mean is calculated for these two anomaly scores using preset weights (0.6 for the first anomaly score and 0.4 for the second anomaly score) to produce a final composite anomaly score, which is used for subsequent health status decisions.

[0042] Among them, when constructing a binary classifier for state classification, a support vector machine model is used, and its training process is achieved by solving the following optimization problem: ; Constraints must be met: ; Among them, w is the normal vector of the hyperplane, is a weight vector to be determined, b is the bias term, is the square of the L2 norm of w, i represents the i-th training sample, n is the total number of training samples, is a slack variable, which is a variable greater than or equal to 0 and is used to measure the degree to which the i-th sample is allowed to violate the classification interval constraint. i is the true category label of the i-th training sample, x i is the original feature vector of the i-th training sample, is the kernel function mapping, and T represents the transpose.

[0043] in, Represents the sample feature vector mapped to the high-dimensional feature space by the kernel function. This implementation uses the radial basis function as the kernel function, which is defined as: ; In the specific implementation, the key parameters are set as follows: penalty factor C = 1.5, scale parameter of radial basis kernel function =0.8. To address the data imbalance problem, a weight coefficient of 2 is set for abnormal class samples during the model training phase, so that they obtain a higher misclassification cost during the optimization process, thereby improving the detection ability of rare abnormal states.

[0044] Step S50: dynamically updating the threshold for determining abnormal status based on the latest acquired comprehensive abnormality score sequence and historical abnormality score data, wherein the threshold can adaptively reflect the statistical distribution characteristics of the device operating status; In this embodiment, based on the latest acquired anomaly score sequence and historical anomaly score data, the threshold for determining abnormal status is dynamically updated as follows: Step S51, continuously obtaining the comprehensive anomaly score generated by real-time calculation to form the latest anomaly score sequence; Step S52: fusing the latest abnormality score sequence with the historically stored normal operating condition abnormality score data to generate an updated abnormality score data set; Step S53, calculating the statistical mean and statistical standard deviation of the anomaly score based on the updated anomaly score data set; In step S54 , based on the calculated statistical mean and statistical standard deviation, combined with the current running time and load status, the threshold generation calculation is performed again to update the threshold.

[0045] In step S51, the system continuously monitors and captures the device's comprehensive anomaly scores generated in real time by the machine learning model. These scores, generated in chronological order, are appended to a predefined storage structure, forming a continuously updated sequence of the latest anomaly scores that reflects the device's recent operating status. This sequence is typically managed as a sliding time window of fixed length (for example, encompassing the last N hours or M sampling points) to ensure that it represents the current status without being overly lengthy.

[0046] In step S52, to ensure stability and historical representativeness of threshold updates, the system fuses the latest anomaly score sequence composed of real-time data with the large-scale anomaly score data stored in the historical database, which has been accumulated over a long period of time under known normal operating conditions. A specific fusion method can be to take the union of these two sets of data to form an updated anomaly score dataset for threshold calculation. To ensure the validity of the historical data, a time decay factor can be applied to the historical data, or only a subset of historical data that matches the current operating conditions (e.g., load, speed) can be selected for fusion.

[0047] Based on the updated anomaly score dataset generated in step S52, step S53 performs the core statistical calculation. The system traverses all anomaly score samples in the dataset and calculates their arithmetic mean and standard.

[0048] The calculated mean and standard deviation jointly describe the central tendency and dispersion of the abnormal scores of the device under normal conditions, providing an accurate statistical basis for the dynamic generation of subsequent thresholds.

[0049] Step S54, based on the calculated statistical mean and statistical standard deviation, combined with the current running time and load status, performs the threshold generation calculation again to update the threshold, including: Step S541, obtaining the calculated updated statistical mean and updated statistical standard deviation; Step S542, obtaining a load sensitivity factor corresponding to the accumulated running time of the current device, wherein the value of the load sensitivity factor decreases according to a predetermined rule as the running time increases; Step S543, obtaining a load deviation coefficient representing the difference between the current operating load state and the reference load state; Step S544 : Add the updated statistical mean, the updated statistical standard deviation of a preset multiple, the product of the load sensitivity factor and the load deviation coefficient to generate an updated adaptive threshold.

[0050] In this embodiment, the statistical mean and statistical standard deviation based on the latest data set, which are recalculated by the previous steps, are first obtained. Subsequently, the system queries and obtains the corresponding load sensitivity factor k from a predefined time-sensitive factor comparison table or function F(T) based on the current cumulative operating time T of the device. The original intention of designing this factor is to reflect the characteristic that the tolerance of the equipment to load fluctuations decreases as the operating time increases and the components gradually age. Therefore, its value follows a predetermined rule (such as linear or exponential decay) and decreases with the increase of operating time T. Next, the system obtains the actual operating load of the current equipment by querying the real-time monitoring system, and compares it with a preset benchmark load to calculate a load deviation coefficient. , whose value is given by the formula =|L current -L base | / L base Determine to quantify the degree of deviation of the current load state from the ideal reference state, where L current is the actual operating load of the current device, L base The preset reference load.

[0051] Finally, the system substitutes all the above parameters into the following formula for synthesis operation: Updated adaptive threshold The core idea of ​​this formula is that the dynamic threshold is composed of the mean value representing the normal score concentration position, 3 times the standard deviation representing the normal fluctuation range, and a compensation term for fine-tuning according to the aging degree of the equipment and the immediate load conditions. Together, they realize the dual adaptation of the threshold to the evolution of the device's own state and external working conditions.

[0052] in, is the standard deviation of abnormal scores of normal samples, is the mean of the abnormal scores of normal samples.

[0053] Step S60, comparing the currently obtained abnormality score with the threshold, and determining whether the device is in an abnormal state based on the comparison result; If it is determined to be abnormal, an alarm signal is generated and maintenance suggestions are output.

[0054] Specifically, the comprehensive anomaly score calculated in real time is numerically compared with the current threshold; If the comprehensive abnormality score is greater than the threshold, generating a first determination signal indicating that the health status is abnormal; If the comprehensive abnormality score is less than or equal to the threshold, generating a second determination signal indicating that the health status is normal; Based on the generated determination signal, the final determination result of the real-time health status of the device is output.

[0055] If the final determination result is that the health status is abnormal, an alarm signal with a predetermined level is generated; Based on the numerical value of the comprehensive abnormality score and the degree of degradation of a specific indicator in the acoustic feature, selecting and generating a corresponding maintenance suggestion from a predetermined set of maintenance measures; Output the alarm signal and the maintenance suggestion.

[0056] In this embodiment, the system compares the comprehensive anomaly score calculated in real time with the currently effective adaptive dynamic threshold, and performs device status determination and decision output based on the comparison result. In specific implementation, the system continuously reads the latest comprehensive anomaly score output by the machine learning model, and compares it in real time with the current threshold determined by the aforementioned dynamic update process. If the value of the comprehensive anomaly score is strictly greater than the current threshold, the system generates a first determination signal indicating that the health status of the device is abnormal; conversely, if the score is less than or equal to the threshold, a second determination signal is generated indicating that the health status is normal. Based on this determination signal, the system finally determines and outputs the determination conclusion of the real-time health status of the device.

[0057] If the final determination is "abnormal health status," the system immediately activates the alarm generation process. This process first determines the predefined alarm level based on the magnitude by which the comprehensive anomaly score exceeds the threshold (for example, exceeding 20% ​​is "warning" and exceeding 50% is "critical") and generates an alarm signal of the corresponding level. Simultaneously, the system initiates the maintenance recommendation generation module. This module not only considers the absolute value of the anomaly score but also analyzes the degree of degradation of specific indicators within the underlying acoustic features that comprise the comprehensive score. For example, a significant increase in spectral energy in a specific high-frequency band may indicate an early bearing failure, while an abnormal increase in a characteristic frequency component in the envelope spectrum may indicate localized damage to the gearbox. The system matches the degradation patterns of these characteristic indicators with a predefined knowledge base containing various maintenance actions (such as "recommend lubrication within one week," "recommend immediate shutdown for bearing inspection," and "recommend gearbox alignment inspection") and selects one or more maintenance recommendations that best match the current fault symptoms. Finally, the system outputs the generated alarm signal and maintenance recommendations to the human-machine interface or higher-level monitoring system, completing the closed loop from condition awareness to decision support.

[0058] In another embodiment of the present invention, if an abnormality is determined, an alarm signal is generated and a maintenance suggestion is output, and the method further includes: If the health status is determined to be abnormal, the abnormality explainability analysis process is started; Based on the comprehensive anomaly score and the acoustic feature, analyzing and determining the contribution of each sub-feature constituting the acoustic feature to the comprehensive anomaly score; Sort the sub-features according to the contribution, and identify one or more sub-features with the highest contribution as the key features that cause the abnormal health status; Mapping and matching the key features with a pre-stored fault feature knowledge base, wherein the fault feature knowledge base records the corresponding relationships between different key features and potential fault causes and maintenance operations; Based on the mapping matching result, generating and outputting diagnostic information for indicating potential fault causes; The diagnostic information is output together with the alarm signal and maintenance recommendation.

[0059] When the present invention is implemented, if it is determined that the health status is abnormal, the system will automatically start the abnormality explainability analysis process. The process first obtains the acoustic feature vector that currently generates a high abnormality score and its corresponding comprehensive abnormality score value, and then the system analyzes the score result through a mechanism based on feature perturbation sensitivity analysis. Specifically, the system will simulate the changes of each sub-item feature value in the feature vector within its possible value range in turn, and keep all other sub-item feature values ​​fixed in each simulation; after each simulated change, the system will input the newly constructed feature vector into the trained machine learning model again to obtain a simulated new abnormality score; by calculating the difference between the new score and the original actual score, the system can quantify the influence of the currently disturbed sub-item feature on the final comprehensive abnormality score; after all sub-item features have undergone the above simulation analysis, the system will normalize the influence of all features and finally obtain the contribution percentage of each sub-item feature to this abnormal event, thereby completing the process from black box model output to transparent contribution analysis.

[0060] After determining the contribution of all sub-features, the system ranks them from highest to lowest based on their contribution values. The top-ranked sub-features, or the ones with the highest contribution, are identified as the key features that caused the abnormal health status. These key features often directly correspond to abnormal conditions in specific physical components within the device. For example, the highest contribution is determined to come from the sub-feature representing meshing impact strength.

[0061] Next, the system maps and matches the identified key features or features with a pre-stored fault feature knowledge base. This knowledge base is a structured database pre-built through domain expert experience and historical fault case data. It details the correspondence between different key features, key feature combination patterns, potential fault causes, and recommended maintenance operations. For example, the knowledge base may define a rule: If "sideband energy ratio" and "high-frequency energy ratio" are both identified as key features, the mapped potential fault cause is "early bearing wear", and the corresponding recommended maintenance operation is "lubricate the bearing and recommend a vibration review within one week."

[0062] Based on the current mapping results, the system automatically generates a diagnostic explanatory message that identifies the potential cause of the fault. This message goes beyond a simple alarm and includes a professional interpretation of key features, an inferred fault mechanism, and a detailed description of potentially affected components, resulting in a diagnostic conclusion with direct guidance.

[0063] Ultimately, the system integrates this diagnostic information with the initial alarm signal and the maintenance recommendations generated based on the ratings, and outputs them to maintenance personnel through channels such as the human-machine interface, the factory's manufacturing execution system, or a remote operation and maintenance platform. This allows maintenance personnel to receive more than just an "abnormal" alarm. Instead, they receive a comprehensive report that includes a preliminary fault diagnosis, decision-making basis, and targeted treatment recommendations, greatly improving the efficiency and accuracy of maintenance decisions.

[0064] More specifically, based on the comprehensive anomaly score and the acoustic features, the contribution of each sub-feature constituting the acoustic features to the comprehensive anomaly score is analyzed and determined by: Obtaining the acoustic feature vector that currently causes abnormal health status, which is composed of multiple sub-item feature values; Simulating in sequence that each sub-item characteristic value in the acoustic characteristic vector changes within its value range while keeping other sub-item characteristic values ​​unchanged; After each simulated change, the changed new feature vector is input into the trained machine learning model to obtain its corresponding new comprehensive anomaly score; By comparing the difference between the new comprehensive anomaly score and the current actual comprehensive anomaly score, the influence of the sub-item features corresponding to the simulation on the comprehensive anomaly score is quantified; The influence degrees corresponding to all sub-item features are normalized, and finally the contribution of each sub-item feature to the current comprehensive anomaly score is obtained.

[0065] Based on the comprehensive anomaly score and the acoustic features, the contribution of each sub-feature of the acoustic features to the comprehensive anomaly score is analyzed and determined through a systematic feature perturbation and sensitivity analysis process. The system first obtains the acoustic feature vector corresponding to the current abnormal state. This vector is composed of multiple pre-defined sub-feature values, such as the specific values ​​of the three sub-features of meshing impact strength, sideband energy ratio, and high-frequency energy ratio.

[0066] The system then enters a loop: for each sub-feature in the feature vector, it simulates a representative change within its possible range of values. A typical simulation method replaces the feature value with the statistical mean of the feature under historical normal conditions. While simulating the change in the current sub-feature, the system maintains the values ​​of all other sub-features in the feature vector unchanged to isolate the impact of a single feature. After each simulated change, the system uses the resulting new feature vector as a new input and feeds it back into the previously trained machine learning model, obtaining a new comprehensive anomaly score for the simulated vector.

[0067] The system then calculates the absolute difference or relative rate of change between this new comprehensive anomaly score and the current actual anomaly score to precisely quantify the impact of the sub-feature corresponding to the simulation operation on the final comprehensive anomaly score result. This impact is an absolute value that represents its sensitivity to change. After all sub-features have completed the above simulation, re-scoring, and difference calculation process, the system obtains the original impact value set corresponding to each feature. Finally, the system uses the softmax function or minimum-maximum normalization method to normalize all the original impact value values ​​so that the sum of the contributions of all features is 100%, thereby ultimately outputting a set of clear and comparable contribution percentages, each percentage representing the contribution of a sub-feature to the cause of the current anomaly score.

[0068] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art will understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of the present invention.

[0069] A system for determining the operating status of a device according to a second embodiment of the present invention is based on the method for determining the operating status of a device according to the first embodiment, and the system includes: a signal acquisition module configured to collect raw acoustic signals from the acoustic sensor; a preprocessing module configured to preprocess the original acoustic signal, including filtering and noise reduction operations, to extract the target frequency band signal and suppress environmental noise; a feature extraction module configured to extract a set of predefined acoustic features from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features; an anomaly score calculation module configured to input the acoustic features into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions; A threshold update module is configured to dynamically update the threshold for determining abnormal status based on the latest acquired comprehensive abnormality score sequence and historical abnormality score data, wherein the threshold can adaptively reflect the statistical distribution characteristics of the device operating status; a state judgment module configured to compare the currently obtained abnormality score with the threshold value and judge whether the device is in an abnormal state according to the comparison result; The maintenance module is configured to generate an alarm signal and output maintenance suggestions if it is determined to be abnormal.

[0070] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0071] It should be noted that the above embodiment provides a system for determining the operating status of a device, which is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps and are not considered to be improper limitations of the present invention.

[0072] An electronic device according to a third embodiment of the present invention includes: at least one processor; and a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned method for determining the operating status of a device.

[0073] A fourth embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are configured to be executed by the computer to implement the above-mentioned method for determining the operating status of a device.

[0074] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes and related instructions of the storage device and processing device described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0075] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0076] The terms "first", "second", etc. are used to distinguish similar objects, rather than to describe or indicate a particular order or sequence.

[0077] The term "comprise" or any other similar term is 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 inherent to such process, method, article, or apparatus.

[0078] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A method for determining the operating status of a device, characterized in that: include: Collecting raw acoustic signals from acoustic sensors; Preprocessing the original acoustic signal, including filtering and noise reduction operations, to extract the target frequency band signal and suppress environmental noise; Extracting a set of predefined acoustic features from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features; Inputting the acoustic features into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions; Dynamically update the threshold for determining abnormal status based on the latest acquired comprehensive abnormality score sequence and historical abnormality score data. The threshold can adaptively reflect the statistical distribution characteristics of the equipment operation status. Comparing the currently obtained abnormality score with the threshold, and determining whether the device is in an abnormal state based on the comparison result; If it is determined to be abnormal, an alarm signal is generated and maintenance suggestions are output.

2. A method for determining the operating status of a device according to claim 1, characterized in that: The original acoustic signal from the acoustic sensor is collected by: A vibration sensor is installed on the surface of the device housing, a reference microphone is deployed in the acoustic environment of the device, and signals from the vibration sensor and the reference microphone are synchronously collected as original acoustic signals.

3. A method for determining the operating status of a device according to claim 1, characterized in that: A set of predefined acoustic features is extracted from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features, and the method is as follows: Perform envelope analysis on the preprocessed signal to obtain its envelope spectrum; Calculating the maximum value of the amplitude within a predetermined frequency range of the envelope spectrum to characterize the intensity of a specific mechanical shock; wherein the specific mechanical shock at least includes a gear meshing shock; Calculating the ratio of the energy of the envelope spectrum within a preset sideband frequency range to the total energy of the envelope spectrum within the full frequency band to indicate the significance of the modulation effect; Calculating the ratio of the energy of the original spectrum in a preset high frequency band to the total energy of the original spectrum in the full frequency band to represent the energy proportion of the high frequency friction or impact noise; The calculated maximum values, sideband energy ratios, and high frequency energy ratios are combined into a set of predefined acoustic eigenvectors.

4. A method for determining the operating status of a device according to claim 1, characterized in that: The acoustic features are input into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions, and the method is as follows: After the equipment is started up and is confirmed to be in a trouble-free operating state, the acoustic characteristic data is continuously collected during a predetermined learning phase; When the amount of the collected fault-free acoustic feature data reaches a preset minimum quantity requirement, the data is output as a data set, and the data set is used to train a support vector machine model and a Gaussian mixture model; The acoustic features extracted in the real-time operation are respectively input into the support vector machine model and the Gaussian mixture model to calculate the corresponding first anomaly score and second anomaly score; A weighted fusion calculation is performed on the first anomaly score and the second anomaly score to obtain a comprehensive anomaly score.

5. A method for determining the operating status of a device according to claim 1, characterized in that: Based on the latest acquired anomaly score sequence and historical anomaly score data, the threshold for judging abnormal status is dynamically updated. The method is as follows: Continuously obtain the comprehensive anomaly scores generated by real-time calculations to form the latest anomaly score sequence; Merging the latest anomaly score sequence with historically stored normal operating condition anomaly score data to generate an updated anomaly score dataset; Calculating the statistical mean and statistical standard deviation of the anomaly scores based on the updated anomaly score dataset; Based on the calculated statistical mean and statistical standard deviation, combined with the current running time and load status, the threshold generation calculation is performed again to update the threshold.

6. A method for determining the operating status of a device according to claim 5, characterized in that: Based on the calculated statistical mean and statistical standard deviation, combined with the current running time and load status, the threshold generation calculation is performed again. The method is: Obtain the calculated updated statistical mean and updated statistical standard deviation; Obtaining a load sensitivity factor corresponding to the accumulated running time of the current device, wherein the value of the load sensitivity factor decreases according to a predetermined rule as the running time increases; Obtaining a load deviation coefficient representing a difference between a current operating load state and a reference load state; The updated statistical mean, the updated statistical standard deviation of a preset multiple, the product of the load sensitivity factor and the load deviation coefficient are added together to generate an updated threshold value.

7. A method for determining the operating status of a device according to claim 1, characterized in that: Compare the currently obtained abnormality score with the threshold, and determine whether the device is in an abnormal state based on the comparison result, the method is as follows: Compare the comprehensive anomaly score calculated in real time with the current threshold; If the comprehensive abnormality score is greater than the threshold, generating a first determination signal indicating that the health status is abnormal; If the comprehensive abnormality score is less than or equal to the threshold, generating a second determination signal indicating that the health status is normal; Based on the generated determination signal, the final determination result of the real-time health status of the device is output.

8. A method for determining the operating status of a device according to claim 7, characterized in that: If it is determined to be abnormal, an alarm signal is generated and maintenance suggestions are output. The method is as follows: If the final determination result is that the health status is abnormal, an alarm signal with a predetermined level is generated; Based on the numerical value of the comprehensive abnormality score and the degree of degradation of a specific indicator in the acoustic feature, selecting and generating a corresponding maintenance suggestion from a predetermined set of maintenance measures; Output the alarm signal and the maintenance suggestion.

9. A system for determining the operating status of a device, based on the method for determining the operating status of a device according to any one of claims 1 to 8, characterized in that: The system includes: a signal acquisition module configured to collect raw acoustic signals from the acoustic sensor; a preprocessing module configured to preprocess the original acoustic signal, including filtering and noise reduction operations, to extract target frequency band signals and suppress environmental noise; a feature extraction module configured to extract a set of predefined acoustic features from the preprocessed signal, wherein the acoustic features include envelope spectrum features, spectrum energy distribution features, and modulation features; an anomaly score calculation module configured to input the acoustic features into a trained machine learning model to obtain a comprehensive anomaly score; the machine learning model is trained based on historical acoustic data under normal operating conditions; A threshold update module is configured to dynamically update the threshold for determining abnormal status based on the latest acquired comprehensive abnormality score sequence and historical abnormality score data, wherein the threshold can adaptively reflect the statistical distribution characteristics of the device operating status; a state judgment module configured to compare the currently obtained abnormality score with the threshold value and judge whether the device is in an abnormal state according to the comparison result; The maintenance module is configured to generate an alarm signal and output maintenance suggestions if it is determined to be abnormal.

10. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the method for determining the operating status of a device according to any one of claims 1 to 8.

Images