A device state monitoring method and system based on vibration signals
By combining adaptive variational mode decomposition and sparse decomposition techniques with a time-series fusion prediction network, the accuracy problem of equipment status monitoring under non-stationary operating conditions is solved, and real-time health status monitoring and remaining life prediction of equipment are realized.
Patent Information
- Application Number
- CN202610502327.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-16
- Publication Date
- 2026-07-03
Smart Images
Figure CN122329642A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mechanical equipment life prediction technology, and in particular to a method and system for monitoring equipment condition based on vibration signals. Background Technology
[0002] During the operation of mechanical equipment, condition monitoring and remaining life prediction are crucial for ensuring production safety and reducing maintenance costs. Vibration monitoring is currently the most widely used sensing method for condition monitoring of industrial equipment. Mechanical vibration transmission loss in rigid bodies is relatively small, making it highly sensitive to responses caused by equipment anomalies or faults, especially for early-stage faults, which can be identified through weak early features in the vibration signal. However, most mechanical equipment exhibits significant time-varying characteristics during operation. Taking a traditional mixer as an example, the material viscosity changes with mixing time, the liquid level gradually decreases, and impact loads are generated during the feeding process. When speed, load, and environmental factors change over time, the mechanical equipment exhibits dynamic behavior, leading to fluctuations in the vibration signal, making it difficult to detect faults using traditional methods that assume a steady state. Under non-stationary operating conditions, the degradation trajectory of the equipment's healthy state is highly nonlinear and time-varying, making it difficult for traditional models to accurately fit the complete degradation process from a healthy state to a faulty state.
[0003] Therefore, developing a device status monitoring method and system that can integrate multimodal signals, adaptively process non-stationary data, accurately identify degradation stages, and make real-time predictions has become an urgent technical problem to be solved in this field. Summary of the Invention
[0004] In view of the problems mentioned above, this invention is proposed. The purpose of this invention is to provide a technical solution to the problems of predicting the remaining life of equipment, analyzing the trend of health degradation, and predicting multimodal feature fusion under non-stationary operating conditions.
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0006] In a first aspect, the present invention provides a device status monitoring method based on vibration signals, comprising the following specific steps:
[0007] S1. Vibration and temperature signals are collected synchronously by sensing units deployed in key parts of the equipment to be monitored;
[0008] S2. Analyze and process the collected signals to obtain the operating status of the equipment and predict potential equipment failures;
[0009] S3. An adaptive variational mode decomposition method based on swarm intelligence optimization algorithm is used to decompose and reconstruct the vibration signal, and a set of degradation-sensitive features is extracted from the reconstructed signal.
[0010] S4. Estimate the instantaneous rotational speed based on the vibration signal itself, and convert the time domain signal into the angular domain signal by equal-angle resampling to eliminate the influence of rotational speed fluctuations on degradation characteristics;
[0011] S5. Construct an overcomplete adaptive dictionary, perform sparse decomposition on the angular domain signal, extract the energy proportion of the sparse coefficient vector as an early degradation sensitivity index, and automatically determine the first prediction time based on this index.
[0012] S6. Integrating vibration and temperature characteristics to construct a normalized health index. ;
[0013] S7. Input the multimodal fusion feature sequence into the temporal fusion prediction network to predict the remaining lifespan of the device. ;
[0014] S8. Generate predictive maintenance recommendations based on the predicted remaining life expectancy and the trend of health indicator decline;
[0015] S9. Use the actual observed equipment failure time as a new sample to trigger incremental retraining and optimize the prediction model parameters.
[0016] Preferably, in step S2, the process from the start of operation to complete failure of the equipment is divided into four stages: health period, early degradation period, accelerated degradation period, and near-failure period.
[0017] Preferably, the adaptive variational mode decomposition method based on swarm intelligence optimization algorithm in step S3 specifically involves: using the Northern Eagle optimization algorithm, with the minimization of envelope entropy as the fitness function, to automatically search for the optimal number of modes K and penalty parameter for VMD. .
[0018] Preferably, step S5, which automatically determines the first prediction time based on the energy proportion of sparse coefficients, specifically includes: calculating the energy proportion of non-zero coefficients in the sparse coefficient vector within the sliding window. ,when When the device exceeds a preset threshold or shows a monotonically increasing trend for N consecutive windows, it is determined that the device has entered the early stage of degradation, and the current time point is recorded as the first prediction time.
[0019] Preferably, the normalized health indicators in step S6 The results were obtained through a comprehensive evaluation and calculation of the root mean square trend index of vibration, kurtosis trend index, sparse decomposition energy ratio index, and temperature change rate index. As the degradation process progresses, it monotonically decreases from 1 to 0.
[0020] Preferably, the temporal fusion prediction network in step S7 includes:
[0021] One-dimensional convolutional layers are used to extract local spatiotemporal features;
[0022] A bidirectional long short-term memory network layer is used to extract long-term dependent temporal features of the degradation process;
[0023] Convolutional block attention module is used to enhance feature representation at critical degradation moments;
[0024] The regression layer is used to output the predicted remaining lifetime.
[0025] Preferably, step S7 further includes quantizing the trained temporal fusion prediction network into INT8 precision format and deploying it on edge computing nodes to achieve real-time prediction output of remaining lifetime.
[0026] Preferably, the step of generating predictive maintenance recommendations in step S8 includes: based on the remaining useful life... With preset threshold , , The comparison results will output suggestions for normal inspection, spare parts arrangement, downtime maintenance, or emergency maintenance.
[0027] A second aspect of the present invention provides a device status monitoring system based on vibration signals, which uses the above-described method for monitoring, including:
[0028] Sensing unit: Deployed in key parts of the equipment to be monitored, including at least one triaxial vibration sensor and one temperature sensor;
[0029] Edge computing node: It communicates with the sensing unit and has an INT8-quantized remaining lifetime prediction network deployed inside, which is used to perform signal preprocessing, degradation feature extraction, health index calculation and remaining lifetime prediction.
[0030] Data storage module: used to store vibration signal data, health indicator sequences, prediction results, and maintenance records;
[0031] Communication module: Supports industrial Ethernet, 5G, and Wi-Fi communication methods for data interaction with the monitoring center;
[0032] Monitoring Center: Used to receive health indicators and remaining life prediction results uploaded by edge computing nodes, providing visualization of health degradation trends and predictive maintenance decision support.
[0033] Preferably, the sensing unit further includes at least one tachometer; a Hall effect speed sensor is installed at the low-speed shaft end of the equipment to directly acquire the instantaneous rotational speed by collecting key phase pulse signals, thereby performing equal-angle resampling. Compared with the method of inverting rotational speed through vibration signals, this method has higher real-time performance and anti-interference capability, and is more suitable for harsh industrial environments.
[0034] The beneficial effects of this invention are:
[0035] 1. In this invention, adaptive VMD accurately eliminates interference components during the dynamic operation of mechanical equipment based on the fault characteristic energy ratio and kurtosis index, providing a clean rotational frequency baseband signal for instantaneous frequency estimation; it can also be applied to locations where physical speed sensors cannot be installed.
[0036] 2. Order tracking transforms time-domain non-stationary signals into angular-domain steady-state signals, making fault impacts exhibit strict periodicity in the angular domain. Sparse decomposition utilizes this characteristic, using periodic impact atoms to efficiently match fault components. Angular-domain synchronous averaging further enhances the signal-to-noise ratio. The combination of these three technologies overcomes the shortcomings of traditional sparse decomposition, which almost fails under varying speeds, and order tracking, which cannot separate weak faults, enabling the accurate capture of weak features in the early degradation stage.
[0037] 3. A health feature with invariant operating conditions was constructed, greatly improving the generalization ability of the deep learning model. VMD denoising, order tracking, and angular domain synchronous average sparse decomposition enhancement jointly completed the "de-operating condition" processing, uniformly transforming the original time-domain waveforms under different speeds and loads into angular domain fault imprints. This imprint is insensitive to changes in operating conditions and can achieve high generalization diagnosis with only a very small number of samples, breaking through the application bottleneck of deep learning in fault diagnosis under varying operating conditions.
[0038] 4. This invention realizes fully automated monitoring of the entire process from data acquisition, signal processing, feature extraction, life prediction to decision support, significantly improving the real-time performance, accuracy, and predictive maintenance practicality of equipment status monitoring under non-stationary operating conditions. Attached Figure Description
[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0040] Figure 1 This is a flowchart of a device status monitoring method based on vibration signals.
[0041] Figure 2 This is a structural diagram of a vibration signal-based equipment status monitoring system. Detailed Implementation
[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0045] Example 1
[0046] Reference Figure 1 This embodiment provides a device status monitoring method based on vibration signals. Taking a common mixer as an example, the method deploys a sensing unit and includes the following specific steps:
[0047] S1. Vibration and temperature signals are collected synchronously by sensing units deployed in key parts of the equipment to be monitored. Key parts include: motor drive end bearing housing, motor non-drive end bearing housing, reducer input shaft bearing housing, reducer housing side, frame bearing housing, near the lower end of the stirring shaft, tank side wall at the corresponding blade installation height, and base anchor bolts.
[0048] The sensing unit includes: at least one triaxial vibration sensor with a sensitivity of 100mV / g and a frequency range of 0.5Hz-10kHz, and a temperature sensor with an accuracy of ±0.5℃.
[0049] All sensors are driven by a unified synchronous clock pulse, synchronously acquiring vibration signal V(t) and temperature signal T(t) at a sampling frequency of fs=25.6kHz, with a single sampling duration of 10 seconds and a sampling interval of 5 minutes. The acquired data is transmitted to the edge computing node via industrial Ethernet.
[0050] S2. Analyze and process the acquired signals to obtain the equipment operating status and predict potential equipment failures; preprocess the acquired multimodal signals.
[0051] The vibration signal is filtered using a sliding median filter to eliminate transient spikes, and then normalized using Z-score.
[0052]
[0053] in, These are the mean and standard deviation of the sampled signal, respectively.
[0054] The temperature signal is filtered using a first-order low-pass filter.
[0055] The process of equipment operation from start to complete failure is then divided into four health phases: the healthy period, the early degradation phase, the accelerated degradation phase, and the near-failure phase. Specifically:
[0056] The healthy period can be defined as when the equipment has just been put into operation and there is no significant degradation. Early degradation can be defined as the appearance of minor fault characteristics, but with still acceptable performance. The accelerated degradation period can be defined as a period of rapid fault development and significantly increased vibration. Nearing the expiration date can be defined as an event that is about to fail and requires immediate maintenance. The stages are divided based on historical failure data of similar equipment, and health indicator thresholds for each stage are determined through statistical methods. The initial prediction time is defined as the point at which the equipment transitions from the healthy phase to the initial stage of degradation.
[0057] S3. The vibration signal is decomposed and reconstructed using an adaptive variational mode decomposition method based on swarm intelligence optimization algorithm, and the degradation-sensitive feature set is extracted from the reconstructed signal. Specifically, the adaptive variational mode decomposition method based on swarm intelligence optimization algorithm in step S3 involves using the Northern Eagle optimization algorithm, with the minimization of envelope entropy as the fitness function, to automatically search for the optimal number of modes K and penalty parameter for VMD. .
[0058] The VMD mathematical model is as follows:
[0059] Variational mode decomposition decomposes the original signal into K eigenmode components with finite bandwidth. Its constrained variational problem is:
[0060]
[0061] In the formula, This is the k-th intrinsic mode component; The center frequency of the k-th component; Let f be the Dirac function; f be the original input signal; and a quadratic penalty factor is introduced. The problem is transformed into an unconstrained problem and solved using the alternating direction multiplier method.
[0062] The parameter adaptive optimization based on the Northern Eagle NGO optimization algorithm is as follows:
[0063] For the number of modes K and the penalty parameter Optimization is performed; where the number of modes K takes the integer range of 2-10, and the penalty parameter is a real number ranging from 500-5000;
[0064] In the fitness function, the smaller the envelope entropy EE, the purer the decomposed modal components and the more obvious the impact characteristics. Its formula is as follows:
[0065]
[0066] In the formula, This represents the number of sampling points for the modal components. The amplitude of the modal component envelope signal is the i-th value. It represents the probability that the i-th amplitude accounts for the sum of all amplitudes.
[0067] NGO optimization steps:
[0068] Initialize the population (K per group, (As a single individual), population size 30, maximum number of iterations 50.
[0069] Perform VMD decomposition on each individual and calculate the envelope entropy of the first modal component after decomposition as the fitness.
[0070] Simulate the prey recognition and pursuit behavior of the northern goshawk to update individual positions (i.e., update K). ).
[0071] Iterate until convergence, and output the optimal parameters corresponding to the minimum envelope entropy.
[0072] Then, VMD decomposition was performed on the original vibration signal using the optimal parameters to obtain... Each modal component. Calculate the fault eigenenergy ratio for each component:
[0073]
[0074] In the formula, The fault characteristic energy ratio of the k-th modal component; For the k-th modal component The Fourier transform spectrum; Preset fault characteristic frequency bands; The sampling frequency is used; the numerator of the integral represents the energy within the fault band, and the denominator represents the total energy across the entire band. (Selection) The components with a strength greater than 0.3 are reconstructed to obtain the denoised signal. ;from The following degradation-sensitive features were extracted:
[0075] Root mean square (RMS) trend Where N is the signal length, To reconstruct the amplitude of the i-th sampling point of the signal;
[0076] cliff ,in for The mean, The standard deviation is denoted as .
[0077] Peak Factor Trend .
[0078] S4. Based on the instantaneous rotational speed estimated from the vibration signal itself, the time-domain signal is converted into an angular-domain signal through equal-angle resampling to eliminate the influence of rotational speed fluctuations on degradation characteristics; This also includes:
[0079] S41, Instantaneous Frequency Estimation
[0080] The obtained noise reduction signal Perform a short-time Fourier transform using a Hanning window with a window length L=512 and an overlap rate of 75% to obtain the time-frequency spectrum. At each time point t, the search range for frequency conversion candidates is determined. The frequency with the highest internal energy is used as the initial instantaneous frequency. Then, the harmonic matching cost function is used for optimization:
[0081]
[0082] In the formula, The instantaneous switching frequency is the optimized value; h represents the harmonic order, ranging from 1 to 5. This represents the amplitude at time t and frequency h, which is the harmonic of the fundamental frequency f, in the time-frequency spectrum. This indicates choosing the frequency f that maximizes the summation; for Perform median filtering for smoothing.
[0083] S42 corner resampling
[0084] Integrating the instantaneous frequency gives the instantaneous phase:
[0085]
[0086] in, Let be the instantaneous phase at time t; Let τ be the instantaneous rotational frequency, and the integration starts from time 0; set the number of sampling points per revolution N=256, then the equal angular intervals... Calculate the theoretical time points corresponding to each equal-angle sampling point. Cubic spline interpolation is used to transform the time-domain signal Mapping to the angular domain yields the angular domain signal. .
[0087] S43 Corner Domain Synchronous Average
[0088] To enhance the impact of periodic faults, for continuous Synchronously average the angular domain signal over one rotation cycle:
[0089]
[0090] In the formula, This is the angular domain signal after synchronization and averaging. The number of rotational cycles participating in the average (taken as 10), where n is the cycle number; This represents the angular domain signal during the nth rotation cycle.
[0091] In practical engineering applications, the rated operating speed or typical speed range of different equipment, such as mixers, fans, compressors, and motors, can usually be obtained directly by consulting technical documents or product manuals, and these ranges vary depending on the actual operating conditions of the equipment model. To improve the robustness and accuracy of instantaneous speed estimation under conditions without a tachometer, this embodiment allows for the pre-setting of a reasonable speed search range for different equipment; for example, a range of 20% to 30% above and below the rated speed.
[0092] This interval, serving as a priori constraint for instantaneous frequency estimation, effectively eliminates misjudgments of harmonic interference, noise spectral peaks, and non-transition frequency components, ensuring that short-time Fourier transform or harmonic matching search is performed only within the preset speed range. By introducing the device's prior speed range, even without a physical tachometer, the estimated instantaneous speed still has a clear physical reference and engineering interpretability, significantly enhancing the reliability of subsequent order tracking and angular domain analysis.
[0093] S5. Construct a complete adaptive dictionary, perform sparse decomposition on the angular domain signal, extract the energy proportion of the sparse coefficient vector as an early degradation sensitivity index, and automatically determine the first prediction time based on this index; specifically, this includes: calculating the energy proportion of non-zero coefficients in the sparse coefficient vector within the sliding window. ,when When the device exceeds a preset threshold or shows a monotonically increasing trend for N consecutive windows, it is determined that the device has entered the early stage of degradation, and the current time point is recorded as the first prediction time.
[0094] It also includes S51 adaptive overcomplete dictionary construction.
[0095] An initial dictionary was constructed using the Adjustable Q-factor Wavelet Transform (TQWT), with a Q-factor of 3, redundancy r=3, and decomposition level J=20. The dictionary was then optimized and trained using the K-SVD algorithm: corner domain signals collected during historical health periods were used as training samples to iteratively update dictionary atoms, resulting in an adaptive dictionary. .
[0096] S52 Sparse Decomposition
[0097] Angular domain synchronous averaging signal In the dictionary Perform sparse decomposition on the above and solve the following optimization problem:
[0098]
[0099] in, It is a sparse coefficient vector; Representing vectors The number of non-zero elements in the middle; For adaptive overcomplete dictionaries; To allow for reconstruction error tolerance;
[0100] The Orthogonal Matching Pursuit (OMP) algorithm is used to solve the problem, and the iteration stopping condition is that the residual energy is less than 1 / 3. Or the number of non-zero coefficients reaches the preset upper limit of 50
[0101] S53 Energy Ratio and First Prediction Time
[0102] Calculate the energy percentage of non-zero coefficients in the sparse coefficient vector:
[0103]
[0104] in, The proportion of sparse energy; Represents the coefficient vector The set of position indices of non-zero elements; This is the i-th non-zero coefficient value; It is the sum of squares of all elements in the coefficient vector; a sliding window is used, with a window length W = 20 sampling points, and the calculation is performed by sliding 5 points at a time. The sequence; when any of the following conditions are met, the device is determined to have entered the early stage of degradation, and the current time point is recorded as the first prediction time (FPT).
[0105] 1.
[0106] 2. Ten consecutive windows It shows a monotonically increasing trend.
[0107] S6. Integrating vibration and temperature characteristics to construct a normalized health index. ;
[0108] Normalized health indicators in step S6 The results were obtained through a comprehensive evaluation and calculation of the root mean square trend index of vibration, kurtosis trend index, sparse decomposition energy ratio index, and temperature change rate index. As the degradation process progresses, it monotonically decreases from 1 to 0.
[0109] Specifically: perform min-max normalization on each feature sequence:
[0110]
[0111] in, Let be the normalized value of the i-th feature at time t; These are the original eigenvalues; These are the minimum and maximum values of this feature in historical health period data, respectively;
[0112] The formulas for calculating health indicators are as follows:
[0113]
[0114] in, Let be the normalized health index at time t, with a value range of [0,1], where 1 represents complete health and 0 represents failure; The normalized root mean square value gradually increases during the degradation process; The normalized kurtosis is close to 0 under normal conditions and increases during faults. The normalized temperature is stable during the healthy period and rises before a failure. The weight coefficients for each feature can be determined using the Analytic Hierarchy Process (AHP). In this embodiment, the values can be 0.3, 0.25, 0.3, and 0.15, respectively.
[0115] S7. Input the multimodal fusion feature sequence into the temporal fusion prediction network to predict the remaining lifespan of the device. ;
[0116] The temporal fusion prediction network in step S7 includes:
[0117] One-dimensional convolutional layers are used to extract local spatiotemporal features;
[0118] A bidirectional long short-term memory network layer is used to extract long-term dependent temporal features of the degradation process;
[0119] Convolutional block attention module is used to enhance feature representation at critical degradation moments;
[0120] The regression layer is used to output the predicted remaining lifetime.
[0121] Step S7 also includes quantizing the trained temporal fusion prediction network into INT8 precision format and deploying it on edge computing nodes to achieve real-time prediction output of remaining lifetime.
[0122] S8. Generate predictive maintenance recommendations based on the predicted remaining life expectancy and the trend of health indicator decline;
[0123] The step of generating predictive maintenance recommendations in step S8 includes: based on the remaining useful life... With preset threshold , , The comparison results will output suggestions for normal inspection, spare parts arrangement, downtime maintenance, or emergency maintenance.
[0124] Specifically, it can be , , Set to 90 days, 30 days, and 7 days respectively; when outputting the remaining service life. When the remaining service life is greater than 90 days, the output suggestion is "normal inspection". When the lifespan is within the range of 30-90 days, the output recommendation is "Arrange spare parts and develop a maintenance plan"; when the remaining service life is output... If the remaining lifespan is greater than 7 days but less than 30 days, the output suggestion should be "Schedule a shutdown for maintenance soon"; when the remaining lifespan is... If the time is less than or equal to 7 days, the output suggestion is "Emergency maintenance, trigger alarm".
[0125] S9. Use the actual observed equipment failure time as new samples to trigger incremental retraining and optimize the prediction model parameters. Incremental retraining is triggered when the number of new samples reaches 200.
[0126] As a preferred embodiment, a Hall effect speed sensor can be installed at the low-speed shaft end of the equipment. The instantaneous speed is directly obtained by acquiring the key phase pulse signal, and then resampled at equal angles. Compared to the method of inferring speed through vibration signals, this method offers higher real-time performance and anti-interference capabilities, making it more suitable for harsh industrial environments. In scenarios where a speed sensor cannot be installed, the instantaneous speed can be estimated based on the vibration signal itself using the STFT and harmonic matching method. Optionally, data from the Hall effect speed sensor can be used to verify the instantaneous speed estimated based on the vibration signal itself to improve the accuracy of the speed estimation.
[0127] In this embodiment, through adaptive VMD decomposition, speed-free order tracking, sparse decomposition early warning, multimodal health index construction, time-series fusion network remaining lifetime prediction and incremental learning, the accurate identification of equipment degradation stages and real-time prediction of remaining lifetime under non-stationary operating conditions are achieved, which significantly improves the early fault warning capability and the reliability of predictive maintenance decisions.
[0128] Example 2
[0129] like Figure 2 As shown, this embodiment provides a device status monitoring system based on vibration signals, which uses the method in Embodiment 1 for monitoring, including:
[0130] Sensing unit: Deployed in key parts of the equipment to be monitored, including at least one triaxial vibration sensor and one temperature sensor;
[0131] Edge computing node: It communicates with the sensing unit and has an INT8-quantized remaining lifetime prediction network deployed inside, which is used to perform signal preprocessing, degradation feature extraction, health index calculation and remaining lifetime prediction.
[0132] Data storage module: used to store vibration signal data, health indicator sequences, prediction results, and maintenance records;
[0133] Communication module: Supports industrial Ethernet, 5G, and Wi-Fi communication methods for data interaction with the monitoring center;
[0134] Monitoring Center: Used to receive health indicators and remaining life prediction results uploaded by edge computing nodes, providing visualization of health degradation trends and predictive maintenance decision support.
[0135] Preferably, the key parts of the equipment to be monitored include vibration-sensitive points, high-frequency fault points, and vibration transmission path points.
[0136] This system integrates adaptive signal decomposition, speed-free order tracking, sparse decomposition, and lightweight time-series prediction networks by deploying multimodal sensing units and edge computing nodes. It can output equipment health indicators and remaining life prediction results in real time under non-stationary operating conditions and provide hierarchical maintenance suggestions. It realizes fully automated monitoring from data acquisition, feature extraction, life prediction to decision support, effectively improving the real-time performance, accuracy, and predictive maintenance practicality of equipment status monitoring under complex operating conditions.
[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method of device condition monitoring based on vibration signals, characterized by, The specific steps include the following: S1. Vibration and temperature signals are collected synchronously by sensing units deployed in key parts of the equipment to be monitored; S2. Analyze and process the collected signals to obtain the operating status of the equipment and predict potential equipment failures; S3. An adaptive variational mode decomposition method based on swarm intelligence optimization algorithm is used to decompose and reconstruct the vibration signal, and a set of degradation-sensitive features is extracted from the reconstructed signal. S4. Estimate the instantaneous rotational speed based on the vibration signal itself, and convert the time domain signal into the angular domain signal by equal-angle resampling to eliminate the influence of rotational speed fluctuations on degradation characteristics; S5. Construct an overcomplete adaptive dictionary, perform sparse decomposition on the angular domain signal, extract the energy proportion of the sparse coefficient vector as an early degradation sensitivity index, and automatically determine the first prediction time based on this index. S6. Integrating vibration and temperature characteristics to construct a normalized health index. ; S7. Input the multimodal fusion feature sequence into the temporal fusion prediction network to predict the remaining lifespan of the device. ; S8. Generate predictive maintenance recommendations based on the predicted remaining life expectancy and the trend of health indicator decline; S9. Use the actual observed equipment failure time as a new sample to trigger incremental retraining and optimize the prediction model parameters.
2. The equipment status monitoring method based on vibration signals according to claim 1, characterized in that: In step S2, the process from the start of operation to complete failure of the equipment is divided into four stages: health period, early degradation period, accelerated degradation period, and near-failure period.
3. The equipment status monitoring method based on vibration signals according to claim 1, characterized in that: The adaptive variational mode decomposition method based on swarm intelligence optimization algorithm in step S3 is as follows: The Northern Eagle optimization algorithm is used, with the minimization of envelope entropy as the fitness function, to automatically search for the optimal number of modes K and penalty parameters for VMD. .
4. The equipment status monitoring method based on vibration signals according to claim 1, characterized in that: Step S5, which automatically determines the first prediction time based on the energy proportion of sparse coefficients, specifically includes: calculating the energy proportion of non-zero coefficients in the sparse coefficient vector within the sliding window. ,when When the device exceeds a preset threshold or shows a monotonically increasing trend for N consecutive windows, it is determined that the device has entered the early stage of degradation, and the current time point is recorded as the first prediction time.
5. The equipment status monitoring method based on vibration signals according to claim 1, characterized in that: Normalized health indicators in step S6 The results were obtained through a comprehensive evaluation and calculation of the root mean square trend index of vibration, kurtosis trend index, sparse decomposition energy ratio index, and temperature change rate index. As the degradation process progresses, it monotonically decreases from 1 to 0.
6. The equipment status monitoring method based on vibration signals according to claim 1, characterized in that: The temporal fusion prediction network in step S7 includes: One-dimensional convolutional layers are used to extract local spatiotemporal features; A bidirectional long short-term memory network layer is used to extract long-term dependent temporal features of the degradation process; Convolutional block attention module is used to enhance feature representation at critical degradation moments; The regression layer is used to output the predicted remaining lifetime.
7. The equipment status monitoring method based on vibration signals according to claim 6, characterized in that: Step S7 also includes quantizing the trained temporal fusion prediction network into INT8 precision format and deploying it on edge computing nodes to achieve real-time prediction output of remaining lifetime.
8. The equipment status monitoring method based on vibration signals according to claim 1, characterized in that: The step of generating predictive maintenance recommendations in step S8 includes: based on the remaining useful life... With preset threshold , , The comparison results will output suggestions for normal inspection, spare parts arrangement, downtime maintenance, or emergency maintenance.
9. A device status monitoring system based on vibration signals, wherein the monitoring is performed using the method described in any one of claims 1-8, characterized in that, include: Sensing unit: Deployed in key parts of the equipment to be monitored, including at least one triaxial vibration sensor and one temperature sensor; Edge computing node: It communicates with the sensing unit and has an INT8-quantized remaining lifetime prediction network deployed inside, which is used to perform signal preprocessing, degradation feature extraction, health index calculation and remaining lifetime prediction. Data storage module: used to store vibration signal data, health indicator sequences, prediction results, and maintenance records; Communication module: Supports industrial Ethernet, 5G, and Wi-Fi communication methods for data interaction with the monitoring center; Monitoring Center: Used to receive health indicators and remaining life prediction results uploaded by edge computing nodes, providing visualization of health degradation trends and predictive maintenance decision support.
10. The equipment status monitoring system based on vibration signals according to claim 9, characterized in that, The sensing unit also includes at least one tachometer; a Hall speed sensor is installed at the low-speed shaft end of the device to directly obtain the instantaneous speed by collecting the key phase pulse signal, thereby performing equal-angle resampling.