Weight meter remote maintenance and intelligent inspection early warning management method and system

Abstract

The invention provides a remote maintenance and intelligent inspection early warning management method and system for a weight meter, and relates to the technical field of intelligent measurement and monitoring, and the method comprises the steps: arranging a multi-source sensor array on a weighing mechanism to collect detection data, carrying out the three-domain conjoint analysis through an edge calculation unit, and extracting multi-dimensional coupling features; the remote monitoring platform establishes a multi-scale damage evolution prediction model based on deep learning to realize the prediction of microscopic crack propagation and macroscopic fatigue damage; an intelligent material is adopted to realize self-adaptive active compensation, and prediction-compensation-evaluation-optimization closed-loop control is formed. The operation reliability and the metering precision of the weight meter can be improved, and the service life of equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to intelligent measurement and monitoring technologies, and particularly to a method and system for remote maintenance and intelligent inspection and warning management of a weight measuring instrument. Background Art

[0002] As a high-precision measurement device, the weight measuring instrument is widely used in industrial production, scientific research and other fields. Its weighing accuracy and long-term stability are directly related to the reliability of the measurement results in downstream processes. However, during the actual operation process, the weight measuring instrument is easily affected by factors such as mechanical vibration, environmental temperature change, and long-term load fatigue, resulting in microscopic crack propagation and macroscopic fatigue damage in the key components of the device, thereby affecting the device performance and service life. At the same time, traditional inspection and maintenance methods usually rely on manual regular inspections, which not only make it difficult to detect potential problems in a timely manner, but also lead to low equipment repair efficiency.

[0003] In recent years, with the rapid development of Internet of Things technology and intelligent manufacturing, remote monitoring technologies based on multi-source data fusion and edge computing have gradually been applied to high-end measurement devices. However, existing monitoring systems are mostly limited to the analysis of single-dimensional data of device operating states, lacking the ability to deeply mine multi-dimensional coupling characteristics and being difficult to comprehensively evaluate the health status of devices. In addition, existing technologies have not fully utilized the coupling relationship between the microscopic crack propagation rate and the macroscopic fatigue damage degree, making it difficult to accurately predict the failure risk of key components of the device and restricting the efficient implementation of intelligent maintenance strategies.

[0004] Therefore, there is an urgent need to develop a method for remote maintenance and intelligent inspection and warning management of a weight measuring instrument based on multi-source data fusion, deep learning algorithms, and intelligent material active compensation technology. Summary of the Invention

[0005] Embodiments of the present invention provide a method and system for remote maintenance and intelligent inspection and warning management of a weight measuring instrument, which can solve the problems in the prior art.

[0006] In the first aspect of the embodiments of the present invention,

[0007] A method for remote maintenance and intelligent inspection and warning management of a weight measuring instrument is provided, including:

[0008] Setting a multi-source sensor array on the weighing mechanism of the weight measuring instrument to collect detection data, performing a three-domain joint analysis in the time domain - frequency domain - wavelet domain, extracting multi-dimensional coupling characteristics, generating an instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism through a deep fusion model, calculating the working state characteristics of the weighing mechanism, forming a multi-dimensional health feature vector, and transmitting it to a remote monitoring platform;

[0009] The remote monitoring platform receives the multi-dimensional health feature vectors and inputs them into a pre-established multi-scale damage evolution prediction model. Among them, the microscopic prediction sub-module calculates the microscopic crack propagation rate based on fracture mechanics, and the macroscopic prediction sub-module analyzes the macroscopic fatigue damage degree based on the cumulative damage theory. According to the coupling relationship between the microscopic crack propagation rate and the macroscopic fatigue damage degree, the failure risk level of key components is determined, a predictive maintenance strategy is generated, and an adaptive compensation control instruction is generated and sent to the edge computing unit;

[0010] The edge computing unit receives and analyzes the adaptive compensation control instruction, establishes an active compensation control model based on intelligent materials, calculates the optimal ratio of each compensation force and controls the corresponding actuators. The edge computing unit collects the real-time response data during the compensation process, constructs an evaluation index system for the compensation effect, generates an evaluation report on the compensation effect, sends the evaluation report on the compensation effect to the remote monitoring platform, online optimizes the parameters of the multi-scale damage evolution prediction model, and updates the predictive maintenance strategy library at the same time, forming a closed-loop control of prediction-compensation-evaluation-optimization.

[0011] In an alternative embodiment,

[0012] A multi-source sensor array is set on the weighing mechanism of the weight measuring instrument to collect detection data, and a three-domain joint analysis of time domain-frequency domain-wavelet domain is carried out to extract multi-dimensional coupling features. Through a deep fusion model, an instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism are generated, including:

[0013] The wireless ad hoc sensor network is used to transmit the detection data collected by the multi-source sensors to the edge computing unit built in the weighing mechanism. The master-slave clock synchronization mechanism is used to align the time of the detection data, and the sampling frequency difference is compensated through an adaptive jitter compensation data cache queue to obtain time-synchronized detection data;

[0014] The wavelet basis function is used to denoise the detection data. For the temperature data in the denoised detection data, the temperature drift of the strain data is eliminated to obtain the compensated strain data;

[0015] The system natural frequency is extracted from the vibration data in the denoised detection data through the fast Fourier transform, the modal damping ratio is estimated by the half-power bandwidth method, and the modulation frequency characteristics are extracted through the Hilbert transform and envelope spectrum analysis to form a frequency domain feature vector;

[0016] The continuous wavelet transform is carried out on the acceleration data to obtain the time-frequency distribution characteristics. The energy distribution of the wavelet coefficients of the time-frequency distribution characteristics is calculated to obtain the wavelet energy characteristics. The wavelet energy characteristics and the frequency domain feature vector are correlated and analyzed to extract multi-dimensional coupling features;

[0017] A multi-layer convolutional structure with residual connections is used to encode the multi-dimensional coupled features. Through skip connections, multi-scale features are retained and mapped to the target space to obtain target features. A multi-head self-attention mechanism is used to adaptively weight and fuse the target features to obtain fused features. Through deconvolution operations, the fused features and the compensated strain data are reconstructed to generate the instantaneous dynamic load distribution map and stress field distribution map of the weighing mechanism.

[0018] In an alternative embodiment,

[0019] Calculate the working state features of the weighing mechanism to form a multi-dimensional health feature vector, including:

[0020] Establish a deep fusion model containing a convolutional neural network and a recurrent neural network, obtain the measurement data of the weighing mechanism, input the measurement data into the deep fusion model, and generate the instantaneous dynamic load distribution map and stress field distribution map of the weighing mechanism as the data to be processed;

[0021] Based on the data to be processed, perform adaptive mesh division on the weighing mechanism, increase the local mesh density in the area where the stress concentration degree is greater than the preset concentration threshold to obtain mesh elements, and calculate the local strain energy density of the mesh elements. The strain energy distribution characteristics of the overall structure are obtained by accumulating the strain energy densities of each mesh element;

[0022] Adopt a finite element stress reconstruction method based on a stress function to construct a biharmonic stress function, establish a stress component mapping relationship, interpolate the strain data of discrete measurement points to obtain weight coefficients, and reconstruct the stress field distribution map in combination with the strain energy distribution characteristics;

[0023] Combine the reconstructed stress field distribution map with the instantaneous dynamic load distribution map, extract the local stress concentration coefficient, modal vibration intensity index, load distribution non-uniformity index, and cumulative damage degree index, and select the principal components with a preset cumulative contribution rate through principal component analysis to construct an initial multi-dimensional health feature vector;

[0024] Use a long short-term memory network to online update the initial multi-dimensional health feature vector to obtain a state vector, and perform weighted fusion on the state vectors of different time scales through an attention mechanism to obtain the final multi-dimensional health feature vector.

[0025] In an alternative embodiment,

[0026] The remote monitoring platform receives the multi-dimensional health feature vector and inputs it into a pre-established multi-scale damage evolution prediction model. Among them, the microscopic prediction sub-module calculates the microscopic crack propagation rate based on fracture mechanics, and the macroscopic prediction sub-module analyzes the macroscopic fatigue damage degree based on the cumulative damage theory, including:

[0027] Perform multi-scale decomposition on the multi-dimensional health feature vector using wavelet transform to obtain high-frequency component signals and low-frequency component signals;

[0028] Input the high-frequency component signals into the microscopic prediction sub-module, obtain the time-frequency energy spectrum through wavelet transform, extract the change characteristics of the stress intensity factor, establish the mapping relationship of the stress field at the crack tip, and obtain the stress field distribution at the crack tip;

[0029] Based on the stress field distribution at the crack tip, divide the crack area into finite element meshes and perform adaptive encryption to construct a node network containing the topological relationship of mesh nodes; use the stress field distribution at the crack tip, material constitutive parameters, crack propagation direction, and critical stress intensity factor as the node characteristics of the node network; adopt a graph convolutional neural network and an attention mechanism to extract dynamic spatial correlation characteristics;

[0030] Calculate the critical crack propagation length according to the dynamic spatial correlation characteristics; establish a functional relationship between the propagation rate and the stress intensity factor using a crack propagation rate formula considering crack closure effect and overload effect to obtain the microscopic crack propagation rate considering crack closure effect and overload effect;

[0031] Input the low-frequency component signals into the macroscopic prediction sub-module, obtain the stress amplitude spectrum and the distribution characteristics of the number of cycles, input the distribution characteristics of the number of cycles into the long short-term memory network, calculate the damage characteristic weight, and obtain the time series dependence characteristics through the time scale attention mechanism;

[0032] Input the damage characteristic weight and the time series dependence characteristics into the analysis unit based on the cumulative damage theory, and calculate the overall fatigue state of the weighing mechanism using the non-linear cumulative damage theory; combine the overall fatigue state with the fatigue life curve, and calculate the equivalent stress amplitude through the damage equivalence criterion to obtain the macroscopic fatigue damage degree.

[0033] In an alternative embodiment,

[0034] Determine the failure risk level of the key components according to the coupling relationship between the microscopic crack propagation rate and the macroscopic fatigue damage degree, and generate a predictive maintenance strategy including:

[0035] Input the microscopic crack propagation rate into the mapping relationship of fracture mechanics theory, perform integral calculation on the microscopic crack propagation rate to obtain the real-time crack length; calculate the initial microscopic remaining strength based on the real-time crack length and the material fracture toughness; multiply the initial microscopic remaining strength by the crack closure effect correction coefficient to obtain the corrected microscopic remaining strength;

[0036] Perform load interaction correction on the macroscopic fatigue damage degree to obtain the corrected damage amount; calculate the strength degradation rate according to the corrected damage amount; perform time integration calculation on the strength degradation rate to obtain the current macroscopic strength;

[0037] Perform a two-scale damage coupling calculation on the corrected microscopic residual strength and the current macroscopic strength to obtain a dynamic correlation coefficient; establish a comprehensive failure criterion equation, and substitute the dynamic correlation coefficient into the comprehensive failure criterion equation to calculate the comprehensive failure criterion value;

[0038] Divide the comprehensive failure criterion value by the allowable stress of the material to obtain a failure criterion ratio; establish a risk level classification standard, compare the failure criterion ratio with the risk level classification standard to obtain the failure risk level; establish a failure distribution function, substitute the failure risk level into the failure distribution function to calculate the failure probability, and perform a weighted calculation on the failure risk level and the failure probability to obtain the component health state assessment result;

[0039] Collect the operating condition parameters of the equipment to establish a maintenance optimization model, substitute the component health state assessment result and the equipment operating condition parameters into the maintenance optimization model for solution to obtain the optimal maintenance time interval;

[0040] Substitute the optimal maintenance time interval and the equipment operating condition parameters into a pre-constructed predictive maintenance strategy generation model to obtain an initial predictive maintenance strategy; collect the real-time condition data of the equipment in real time, compare the real-time condition data of the equipment with the preset condition range, and when the real-time condition data of the equipment exceeds the preset condition range, substitute the real-time condition data of the equipment into the maintenance optimization model to recalculate and update the predictive maintenance strategy.

[0041] In an alternative embodiment,

[0042] Substituting the optimal maintenance time interval and the equipment operating condition parameters into a pre-constructed predictive maintenance strategy generation model to obtain an initial predictive maintenance strategy includes:

[0043] Divide it into multiple time sub-intervals according to the change rate of the optimal maintenance time interval sequence to obtain a time sub-interval sequence; calculate the mutual information and conditional entropy between the equipment operating condition parameters, and determine the key condition parameter set based on the mutual information and conditional entropy;

[0044] For the time sub-interval sequence, calculate the state embedding features of each time sub-interval and perform weighted fusion to obtain a time feature vector; construct a parameter correlation graph based on the key condition parameter set, and extract the condition feature vector of the parameter correlation graph;

[0045] Input the time feature vector and the condition feature vector into a pre-constructed hierarchical strategy generation network; wherein, the first layer generates a maintenance timing template based on the time feature vector, the second layer generates a maintenance action configuration based on the condition feature vector, and the third layer fuses the maintenance timing template and the maintenance action configuration to generate a candidate maintenance strategy;

[0046] Collect the performance parameters before and after the maintenance of the acquisition device, calculate the recovery degree of the performance parameters after maintenance relative to those before maintenance, and obtain the maintenance effect index; monitor the degradation state of the key components of the device, calculate the failure risk probability based on the degradation rate and the remaining life prediction results, and obtain the reliability index; count the labor cost, spare part cost and downtime loss during the maintenance process, calculate the total maintenance cost per unit time, and obtain the cost index.

[0047] Optimize and adjust the parameters of the network for generating the hierarchical strategy according to the maintenance effect index, reliability index and cost index. When the maintenance effect index is greater than the first preset threshold, the reliability index is greater than the second preset threshold, and the cost index is less than the third preset threshold, determine the current candidate maintenance strategy as the initial predictive maintenance strategy.

[0048] In an alternative embodiment,

[0049] The edge computing unit receives and parses the adaptive compensation control instruction, establishes an active compensation control model based on intelligent materials, and calculates the optimal ratio of each compensation force and controls the corresponding actuator, including:

[0050] Extract the compensation target displacement value, compensation time threshold and compensation accuracy threshold from the adaptive compensation control instruction;

[0051] Collect the data of the environmental temperature sensor to obtain the temperature change value, calculate the thermal expansion displacement value of the piezoelectric material and the thermal expansion displacement value of the magnetostrictive material respectively according to the temperature change value, establish the displacement response equation of the piezoelectric material, and substitute the piezoelectric constant, piezoelectric drive voltage and the thermal expansion displacement value of the piezoelectric material into the displacement response equation of the piezoelectric material to obtain the compensation displacement value of the piezoelectric material; at the same time, calculate the compensation displacement value of the magnetostrictive material according to the magnetostrictive coefficient, magnetostrictive drive magnetic field strength and the thermal expansion displacement value of the magnetostrictive material;

[0052] Establish an intelligent material collaborative compensation mechanism including the piezoelectric material ratio coefficient and the magnetostrictive material ratio coefficient, multiply the compensation displacement value of the piezoelectric material by the piezoelectric material ratio coefficient to obtain the weighted displacement value of the piezoelectric material, and multiply the compensation displacement value of the magnetostrictive material by the magnetostrictive material ratio coefficient to obtain the weighted displacement value of the magnetostrictive material;

[0053] Based on the intelligent material collaborative compensation mechanism, add the weighted displacement value of the piezoelectric material and the weighted displacement value of the magnetostrictive material to obtain the total compensation displacement value; perform nonlinear optimization and solution on the total compensation displacement value, compensation target displacement value, compensation time threshold and compensation accuracy threshold to obtain the optimal piezoelectric material ratio coefficient and the optimal magnetostrictive material ratio coefficient;

[0054] Substitute the optimal piezoelectric material ratio coefficient, the compensation target displacement value, and the piezoelectric material thermal expansion displacement value into the piezoelectric material displacement response equation, and solve for the target piezoelectric drive voltage by inverse solution. At the same time, substitute the optimal magnetostrictive material ratio coefficient, the compensation target displacement value, and the magnetostrictive material thermal expansion displacement value into the magnetostrictive material displacement response equation, and solve for the target magnetic field drive intensity by inverse solution.

[0055] Collect the data of the displacement sensor to obtain the real-time compensation displacement value, and compare the real-time compensation displacement value with the compensation target displacement value to obtain the displacement tracking error value.

[0056] Based on the displacement tracking error value, adjust the target piezoelectric drive voltage and the target magnetic field drive intensity in real time to obtain the compensation drive voltage and the compensation magnetic field intensity; control the piezoelectric material actuator and the magnetostrictive material actuator according to the compensation drive voltage and the compensation magnetic field intensity to achieve the adaptive active compensation of the weighing mechanism.

[0057] In the second aspect of the embodiments of the present invention,

[0058] A remote maintenance and intelligent inspection and warning management system for a weight measuring instrument is provided, including:

[0059] A first unit for setting a multi-source sensor array on the weighing mechanism of the weight measuring instrument to collect detection data, performing three-domain joint analysis in the time domain, frequency domain, and wavelet domain, extracting multi-dimensional coupling features, generating an instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism through a deep fusion model, calculating the working state features of the weighing mechanism, forming a multi-dimensional health feature vector, and transmitting it to the remote monitoring platform;

[0060] A second unit for the remote monitoring platform to receive the multi-dimensional health feature vector and input it into a pre-established multi-scale damage evolution prediction model. Among them, the microscopic prediction sub-module calculates the microscopic crack propagation rate based on fracture mechanics, the macroscopic prediction sub-module analyzes the macroscopic fatigue damage degree based on the cumulative damage theory, determines the failure risk level of key components according to the coupling relationship between the microscopic crack propagation rate and the macroscopic fatigue damage degree, generates a predictive maintenance strategy, and generates an adaptive compensation control instruction, which is sent to the edge computing unit;

[0061] A third unit for the edge computing unit to receive and parse the adaptive compensation control instruction, establish an active compensation control model based on intelligent materials, calculate the optimal ratio of each compensation force and control the corresponding actuator, collect the real-time response data during the compensation process, construct an evaluation index system for the compensation effect to generate a compensation effect evaluation report, online optimize the parameters of the multi-scale damage evolution prediction model based on the compensation effect evaluation report, and at the same time update the predictive maintenance strategy library to form a closed-loop control of prediction-compensation-evaluation-optimization.

[0062] In the third aspect of the embodiments of the present invention,

[0063] Provided is an electronic device, comprising:

[0064] a processor;

[0065] a memory for storing instructions executable by the processor;

[0066] wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0067] In a fourth aspect of the embodiments of the present invention,

[0068] provided is a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0069] In this embodiment, by setting a multi-source sensor array on the weighing mechanism to collect detection data and using an edge computing unit for joint time-domain, frequency-domain, and wavelet-domain analysis, the operating state information of the device can be comprehensively and accurately obtained, realizing real-time monitoring and evaluation of the working state of the weighing mechanism, and improving the accuracy and reliability of state monitoring. By using a deep learning algorithm to establish a multi-scale damage evolution prediction model and combining microscopic crack propagation with macroscopic fatigue damage for analysis, the failure risk of key components of the device can be accurately predicted, potential faults can be discovered in advance, and a scientific and reasonable predictive maintenance strategy can be formulated, effectively reducing the device failure rate and maintenance cost. By establishing an active compensation control model based on intelligent materials, adaptive active compensation of the weighing mechanism is realized, and an evaluation index system for compensation effect is constructed for online optimization, forming a closed-loop control of prediction, compensation, and evaluation optimization, significantly improving the reliability and service life of the device, and ensuring the measurement accuracy and stability of the weight measuring instrument. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a schematic flow chart of the method for remote maintenance and intelligent patrol warning management of the weight measuring instrument according to the embodiments of the present invention;

[0071] Figure 2 is a heat map of the multi-source data denoising and temperature compensation effect according to the embodiments of the present invention;

[0072] Figure 3 is a performance comparison chart of the predictive maintenance strategy generation model according to the embodiments of the present invention;

[0073] Figure 4 is a schematic structural diagram of the remote maintenance and intelligent patrol warning management system of the weight measuring instrument according to the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0075] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0076] Figure 1 It is a schematic flowchart of the remote maintenance and intelligent inspection and warning management method for the weight measuring instrument in the embodiments of the present invention. As Figure 1 shown, the method includes:

[0077] A multi-source sensor array is arranged on the weighing mechanism of the weight measuring instrument to collect detection data, and a three-domain joint analysis in the time domain, frequency domain, and wavelet domain is performed to extract multi-dimensional coupling features. An instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism are generated through a deep fusion model, the working state features of the weighing mechanism are calculated, a multi-dimensional health feature vector is formed, and it is transmitted to the remote monitoring platform;

[0078] The remote monitoring platform receives the multi-dimensional health feature vector and inputs it into a pre-established multi-scale damage evolution prediction model. Among them, the microscopic prediction sub-module calculates the microscopic crack propagation rate based on fracture mechanics, and the macroscopic prediction sub-module analyzes the macroscopic fatigue damage degree based on the cumulative damage theory. According to the coupling relationship between the microscopic crack propagation rate and the macroscopic fatigue damage degree, the failure risk level of key components is determined, a predictive maintenance strategy is generated, and an adaptive compensation control instruction is generated and sent to the edge computing unit;

[0079] The edge computing unit receives and parses the adaptive compensation control instruction, establishes an active compensation control model based on intelligent materials, calculates the optimal ratio of each compensation force and controls the corresponding actuator, collects the real-time response data during the compensation process, constructs an evaluation index system for the compensation effect to generate a compensation effect evaluation report, online optimizes the parameters of the multi-scale damage evolution prediction model based on the compensation effect evaluation report, and at the same time updates the predictive maintenance strategy library to form a closed-loop control of prediction-compensation-evaluation-optimization.

[0080] In an alternative embodiment, arranging a multi-source sensor array on the weighing mechanism of the weight measuring instrument to collect detection data, and performing a three-domain joint analysis in the time domain, frequency domain, and wavelet domain to extract multi-dimensional coupling features, and generating an instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism through a deep fusion model includes:

[0081] The detection data collected by multi-source sensors are transmitted to the edge computing unit built in the weighing mechanism through a wireless ad hoc sensor network. The time of the detection data is aligned by using a master-slave clock synchronization mechanism, and the sampling frequency difference is compensated through an adaptive jitter compensation data cache queue to obtain time-synchronized detection data.

[0082] The detection data are denoised by wavelet basis functions. For the temperature data in the denoised detection data, the temperature drift of the strain data is eliminated to obtain compensated strain data.

[0083] The natural frequency of the system is extracted from the vibration data in the denoised detection data through fast Fourier transform. The modal damping ratio is estimated by the half-power bandwidth method. The modulation frequency characteristics are extracted through Hilbert transform and envelope spectrum analysis to form a frequency domain feature vector.

[0084] The continuous wavelet transform is performed on the acceleration data to obtain time-frequency distribution characteristics. The energy distribution of the wavelet coefficients of the time-frequency distribution characteristics is calculated to obtain wavelet energy characteristics. The wavelet energy characteristics and the frequency domain feature vector are subjected to correlation analysis to extract multi-dimensional coupling characteristics.

[0085] A multi-layer convolutional structure with residual connections is used to encode the multi-dimensional coupling characteristics. The multi-scale characteristics are retained through skip connections and mapped to the target space to obtain target characteristics. The multi-head self-attention mechanism is used to perform adaptive weighted fusion on the target characteristics to obtain fusion characteristics. The fusion characteristics and the compensated strain data are reconstructed through deconvolution operations to generate the instantaneous dynamic load distribution map and stress field distribution map of the weighing mechanism.

[0086] Exemplarily, the dynamic load and stress field distribution detection system of the weighing mechanism first collects multi-source detection data through strain sensors, temperature sensors, vibration sensors, and acceleration sensors arranged at key parts of the weighing mechanism. Each type of sensor uses a wireless ad hoc sensor network for data transmission. The master node in the network periodically sends clock synchronization frames to the slave nodes. After receiving the synchronization frames, the slave nodes calibrate their local clocks according to the timestamp information. Considering the differences in the sampling frequencies of different types of sensors, the system sets an adaptive data cache queue. The data of sensors with higher sampling frequencies are downsampled, and the data of sensors with lower sampling frequencies are interpolated and compensated to achieve time alignment of multi-source data. For example, if the sampling frequency of the strain sensor is 1000Hz and the sampling frequency of the temperature sensor is 10Hz, then linear interpolation is performed on the temperature data to obtain a 1000Hz synchronous data stream.

[0087] Preprocess the original data for time synchronization. Use the db4 wavelet basis function to perform 4-layer wavelet decomposition on the data to obtain wavelet coefficients in different frequency bands. Calculate the optimal soft threshold based on the Bayesian criterion and perform threshold denoising on the wavelet coefficients. Taking the influence of temperature on strain as an example, when the ambient temperature rises from 20°C to 40°C, the output of the strain sensor will produce a drift of about 50 με. Establish a temperature-strain compensation model based on the thermodynamic principle, substitute the denoised temperature data into the model, and eliminate the influence of temperature drift in the strain data.

[0088] Perform a fast Fourier transform on the denoised vibration data to extract the main natural frequencies of the system in the range of 0 - 500 Hz. Calculate the corresponding modal damping ratios using the half-power bandwidth method, with typical values between 0.5% and 5%. Perform a Hilbert transform on the vibration data to obtain the analytic signal, and extract the modulation frequency characteristics through envelope spectrum analysis. Combine the system natural frequencies, modal damping ratios, and modulation frequency characteristics to form a frequency-domain feature vector.

[0089] At the same time, perform a continuous wavelet transform on the denoised acceleration data, select the Mexican hat wavelet as the basis function, and calculate the time-frequency distribution characteristics. Statistically analyze the energy distribution of the wavelet coefficients on the time-frequency plane to obtain the wavelet energy characteristics reflecting the dynamic characteristics of the system. Perform a correlation analysis on the wavelet energy characteristics and the frequency-domain feature vector to extract multi-dimensional coupling characteristics.

[0090] Use a 5-layer convolutional neural network to encode the multi-dimensional coupling characteristics. Add residual connections after each convolution to avoid gradient disappearance. Integrate the feature information of different scales through skip connections and map it to a 128-dimensional target feature space. Set 8 attention heads in the target space, and each attention head calculates the attention weights through 64-dimensional query vectors, key vectors, and value vectors to achieve adaptive fusion of the features. Finally, reconstruct the fused features and the compensated strain data into the instantaneous dynamic load distribution map and stress field distribution map of the weighing mechanism through a 3-layer deconvolution network, with a resolution of 256×256 pixels.

[0091] Figure 2 This is the heat map of the multi-source data denoising and temperature compensation effect of the embodiment of the present invention, as shown in Figure 2As shown, this heat map demonstrates the significant differences in strain data before and after temperature compensation. The heat map of the original strain data on the left shows that within the ambient temperature range of 18.2°C to 32.7°C (a change of 14.5°C), the original strain data collected by the sensor exhibits an obvious temperature drift phenomenon, with the values fluctuating greatly between 673 - 891 με and a standard deviation as high as 52.7 με. The color change in the figure is obvious, showing a gradient distribution from the upper left to the lower right, indicating that the strain data drifts significantly with temperature changes. The heat map on the right shows the strain data after denoising with wavelet basis functions and temperature compensation in this technical solution, indicating that the strain data at each measurement point is basically stable within the range of 480 - 487 με, with a standard deviation of only ±2.5 με and a uniform color distribution. Compared with traditional linear temperature compensation methods (such as the temperature-strain linear regression model), this technical solution can more accurately identify and eliminate the non-linear influence of temperature changes on strain data through wavelet analysis, and the compensation accuracy is improved by approximately 95.3%. This proves that the temperature compensation algorithm adopted in this technical solution can effectively eliminate the influence of temperature drift, providing accurate basic strain data for subsequent load analysis, especially suitable for industrial field application scenarios with large ambient temperature fluctuations.

[0092] In this embodiment, reliable transmission and time synchronization of multi-source heterogeneous data are achieved through a wireless ad-hoc sensor network and an adaptive data caching queue, ensuring the integrity and consistency of the detection data and providing a high-quality data basis for subsequent analysis. Based on multi-domain analysis methods such as wavelet multi-resolution analysis, temperature compensation, and spectrum analysis, noise reduction and feature extraction of the detection data are realized, effectively eliminating the interference of environmental factors and improving the accuracy and robustness of feature expression. A deep learning model is used to adaptively fuse and reconstruct multi-dimensional coupling features, making full use of residual connections and attention mechanisms to capture the correlation between features, and achieving high-precision reconstruction of the dynamic load and stress field distribution of the weighing mechanism, providing a reliable basis for equipment status assessment and early warning.

[0093] In an optional implementation manner, calculate the working state characteristics of the weighing mechanism to form a multi-dimensional health feature vector, including:

[0094] Establish a deep fusion model containing a convolutional neural network and a recurrent neural network, obtain the measurement data of the weighing mechanism, input the measurement data into the deep fusion model, and generate the instantaneous dynamic load distribution map and stress field distribution map of the weighing mechanism as the data to be processed;

[0095] Based on the data to be processed, perform adaptive mesh division on the weighing mechanism, increase the local mesh density in the area where the stress concentration is greater than the preset concentration threshold to obtain grid cells, and calculate the local strain energy density of the grid cells. The strain energy distribution characteristics of the overall structure are obtained by accumulating the strain energy densities of each grid cell;

[0096] A biharmonic stress function is constructed by using a finite element stress reconstruction method based on a stress function, a mapping relationship of stress components is established, and weight coefficients are obtained by interpolating the strain data of discrete measurement points. Combining the characteristics of strain energy distribution, a stress field distribution map is reconstructed.

[0097] The reconstructed stress field distribution map is combined with the instantaneous dynamic load distribution map, and local stress concentration coefficients, modal vibration intensity indexes, load distribution non-uniformity indexes, and cumulative damage degree indexes are extracted. Principal components with a preset cumulative contribution rate are selected through principal component analysis to construct an initial multi-dimensional health feature vector.

[0098] The long short-term memory network is used to online update the initial multi-dimensional health feature vector to obtain a state vector, and the state vectors at different time scales are weighted and fused through an attention mechanism to obtain the final multi-dimensional health feature vector.

[0099] Exemplarily, the method for generating the health feature vector of a weighing mechanism first needs to establish a deep fusion model. This model adopts an architecture combining a convolutional neural network and a recurrent neural network. The convolutional neural network includes multiple convolutional layers and pooling layers for extracting the spatial features of the measurement data of the weighing mechanism. The recurrent neural network adopts a bidirectional structure for capturing temporal features. The input layer of the deep fusion model receives measurement data such as strain, displacement, and acceleration of the weighing mechanism, and outputs an instantaneous dynamic load distribution map and a stress field distribution map after feature extraction.

[0100] When performing adaptive mesh division on the generated distribution map, first determine the grid density distribution function based on the stress gradient. The quadtree algorithm is used to implement grid encryption in the stress concentration area, and the grid size decreases as the stress gradient increases. For each grid cell, the strain field is obtained by interpolating the displacements of the cell nodes, and then the strain energy density of the cell is calculated. Taking a certain weighing sensor as an example, the grid size in the stress concentration area can reach 0.1 mm, while the grid size in the low-stress area is 2 mm, and the total number of grid cells is about 50,000.

[0101] In the stress field reconstruction step, the mapping relationship between stress components and stress functions is established by using the biharmonic stress function method. The strain data of the measurement points are interpolated by the Gaussian radial basis function, and the interpolation radius is taken as 1.5 times the distance between the measurement points. Combining the aforementioned strain energy distribution characteristics, an iterative algorithm is used to reconstruct the stress field, and the iteration termination condition is that the relative error between two adjacent calculation results is less than 0.1%.

[0102] When extracting characteristic indicators, the local stress concentration factor is obtained by calculating the ratio of the maximum local stress to the nominal stress; the modal vibration intensity index considers the first 6 modes, and the sum of the products of the amplitudes of each mode and the corresponding natural frequencies is obtained; the load distribution non-uniformity index is characterized by calculating the ratio of the standard deviation to the mean of the load values of the grid elements; the cumulative damage degree index is calculated based on the stress cycle times under actual working conditions and the fatigue life determined by the material S-N curve.

[0103] When performing principal component analysis on the extracted characteristic indicators, by calculating the eigenvalues and eigenvectors, the principal components with a cumulative contribution rate reaching 95% are selected to construct the initial health characteristic vector. The long short-term memory network is used to update the characteristic vector online. The network includes three control units: the forgetting gate, the input gate, and the output gate, and the number of hidden layer nodes is set to 128. The state vectors at different time scales are weighted and fused through the attention mechanism, and the attention weights are calculated through the softmax function, and finally a multi-dimensional characteristic vector reflecting the health state of the weighing mechanism is generated.

[0104] In this embodiment, the high-precision reconstruction of the instantaneous dynamic load and stress field distribution of the weighing mechanism is realized through the deep fusion model, which overcomes the limitation that it is difficult for traditional methods to obtain load and stress information simultaneously, and lays a foundation for health characteristic extraction. Based on the adaptive grid division and stress reconstruction technology, the refined analysis of the stress concentration area is realized, the stress field reconstruction accuracy is improved, the characteristic extraction is more accurate and reliable, and the local damage state of the weighing mechanism can be effectively reflected. The long short-term memory network and the attention mechanism are used to update the health characteristic vector online and perform multi-scale fusion, which improves the characterization ability of the characteristic vector for the dynamic characteristics and damage evolution law of the weighing mechanism, and provides a reliable basis for subsequent health state assessment and life prediction.

[0105] In an alternative embodiment, the remote monitoring platform receives the multi-dimensional health characteristic vector and inputs it into a pre-established multi-scale damage evolution prediction model. The microscopic prediction sub-module calculates the microscopic crack propagation rate based on fracture mechanics, and the macroscopic prediction sub-module analyzes the macroscopic fatigue damage degree based on the cumulative damage theory, including:

[0106] The multi-dimensional health characteristic vector is decomposed into multi-scales by wavelet transform to obtain high-frequency component signals and low-frequency component signals;

[0107] The high-frequency component signal is input into the microscopic prediction sub-module. Through wavelet transform, the time-frequency energy spectrum is obtained, the change characteristics of the stress intensity factor are extracted, the mapping relationship of the stress field at the crack tip is established, and the stress field distribution at the crack tip is obtained;

[0108] Based on the stress field distribution at the crack tip, the crack region is divided into finite element meshes and adaptively refined, and a node network including the topological relationship of mesh nodes is constructed; the stress field distribution at the crack tip, material constitutive parameters, crack propagation direction, and critical stress intensity factor are used as node features of the node network; a graph convolutional neural network and an attention mechanism are used to extract dynamic spatial correlation features;

[0109] The critical crack propagation length is calculated according to the dynamic spatial correlation features; a function relationship between the propagation rate and the stress intensity factor is established using a crack propagation rate formula considering crack closure effect and overload effect to obtain the microscopic crack propagation rate considering crack closure effect and overload effect;

[0110] The low-frequency component signal is input into the macroscopic prediction sub-module to obtain the stress amplitude spectrum and the cycle number distribution characteristics. The cycle number distribution characteristics are input into the long short-term memory network to calculate the damage feature weights, and the time series dependence features are obtained through the time scale attention mechanism;

[0111] The damage feature weights and the time series dependence features are input into the analysis unit based on the cumulative damage theory, and the overall fatigue state of the weighing mechanism is calculated using the nonlinear cumulative damage theory; the overall fatigue state is combined with the fatigue life curve, and the equivalent stress amplitude is calculated through the damage equivalence criterion to obtain the macroscopic fatigue damage degree.

[0112] Exemplarily, the multi-scale damage evolution prediction model first processes the input multi-dimensional health feature vector. Through wavelet transform for multi-scale decomposition, using the db4 wavelet basis function, the decomposition level is set to 4 layers, obtaining high-frequency component signals and low-frequency component signals. The high-frequency component signals mainly contain information related to microscopic crack propagation, while the low-frequency component signals reflect macroscopic fatigue damage characteristics.

[0113] For the microscopic prediction sub-module, after inputting the high-frequency component signal, continuous wavelet transform is used to obtain the time-frequency energy spectrum. The Morlet wavelet is selected as the mother wavelet function, the scale parameter range is set from 1 to 128, and the time displacement step size is 1. By analyzing the peak distribution of the energy spectrum, the change characteristics of the stress intensity factor are extracted. Based on the extracted features, a stress field mapping relationship at the crack tip is established, and the stress field distribution at the crack tip is obtained by applying the stress field reconstruction algorithm.

[0114] After obtaining the stress field distribution, an adaptive mesh generation technique is used to mesh the crack region. In the crack tip region, the mesh size is set to 0.1 mm and gradually transitions to 1 mm outward. A network including the node topological relationship is constructed, and each node contains information such as position coordinates, stress components, and displacement components. The node features include the stress field distribution, material elastic modulus and Poisson's ratio, crack propagation direction angle, and critical stress intensity factor value.

[0115] Design a multi-layer graph convolutional neural network to process node features. The network consists of 3 graph convolutional layers, and the output dimensions of each layer are 64, 128, and 256 respectively. The spatial attention layer adopts the dot product attention mechanism, and the temporal attention layer uses the self-attention structure based on LSTM to extract dynamic spatial correlation features.

[0116] The fracture mechanics analysis unit calculates the critical crack propagation length based on the extracted features. Considering the crack closure effect, a closure stress ratio parameter is introduced, and its value range is from 0.2 to 0.4. The overload effect is corrected by introducing a delay coefficient, and the coefficient range is from 0.5 to 0.8. Finally, the corrected crack propagation rate is obtained.

[0117] The macro prediction sub-module receives the low-frequency component signal and performs time-frequency analysis using the short-time Fourier transform. The window length is set to 1024 points. The stress cycle characteristics are statistically analyzed by an improved rainflow counting algorithm, and a stress amplitude correction coefficient is introduced, with a range from 0.8 to 1.2.

[0118] Design a bidirectional gated recurrent unit network to process the distribution characteristics of the number of cycles. The network consists of 2 bidirectional GRU layers, and the hidden layer dimension is 128. The hierarchical attention mechanism includes two levels: feature-level attention and time-scale attention. The feature-level attention layer outputs the damage feature weight vector, and the time-scale attention layer captures the long-term temporal dependence relationship.

[0119] Finally, calculate the overall fatigue state based on the non-linear cumulative damage theory. Through the damage equivalence criterion, establish the corresponding relationship between the equivalent stress amplitude and the fatigue life, and obtain the evaluation result of the macro fatigue damage degree.

[0120] In this embodiment, the collaborative prediction of micro crack propagation and macro fatigue damage is realized through multi-scale decomposition, improving the accuracy and reliability of the prediction model. The use of adaptive mesh division and graph convolutional neural network enhances the accuracy of crack propagation analysis. The introduction of spatial attention and temporal attention mechanisms enhances the model's ability to extract dynamic features. Considering the crack closure effect and overload effect makes the crack propagation prediction more in line with the actual situation. The use of a hierarchical attention structure to process the distribution characteristics of the number of cycles improves the accuracy of macro fatigue damage prediction. Through the non-linear cumulative damage theory and damage equivalence criterion, the accurate evaluation of fatigue life is realized.

[0121] In an alternative embodiment, determine the failure risk level of key components according to the coupling relationship between the micro crack propagation rate and the macro fatigue damage degree. The generated predictive maintenance strategy includes:

[0122] Input the microcrack growth rate into the mapping relationship of fracture mechanics theory, and perform integral calculation on the microcrack growth rate to obtain the real-time crack length; calculate the initial micro-residual strength based on the real-time crack length and the material fracture toughness; multiply the initial micro-residual strength by the crack closure effect correction coefficient to obtain the corrected micro-residual strength;

[0123] Perform load interaction correction on the macroscopic fatigue damage degree to obtain the corrected damage amount; calculate the strength degradation rate based on the corrected damage amount; perform time integral calculation on the strength degradation rate to obtain the current macroscopic strength;

[0124] Perform two-scale damage coupling calculation on the corrected micro-residual strength and the current macroscopic strength to obtain the dynamic correlation coefficient; establish a comprehensive failure criterion equation, and substitute the dynamic correlation coefficient into the comprehensive failure criterion equation to calculate the comprehensive failure criterion value;

[0125] Divide the comprehensive failure criterion value by the allowable stress of the material to obtain the failure criterion ratio; establish a risk level classification standard, compare the failure criterion ratio with the risk level classification standard to obtain the failure risk level; establish a failure distribution function, substitute the failure risk level into the failure distribution function to calculate the failure probability, and perform weighted calculation on the failure risk level and the failure probability to obtain the component health status evaluation result;

[0126] Collect the operating condition parameters of the equipment to establish a maintenance optimization model, substitute the component health status evaluation result and the equipment operating condition parameters into the maintenance optimization model for solution, and obtain the optimal maintenance time interval;

[0127] Substitute the optimal maintenance time interval and the equipment operating condition parameters into the pre-constructed predictive maintenance strategy generation model to obtain the initial predictive maintenance strategy; collect the real-time condition data of the equipment in real time, compare the real-time condition data of the equipment with the preset condition range, and when the real-time condition data of the equipment exceeds the preset condition range, substitute the real-time condition data of the equipment into the maintenance optimization model to recalculate and update the predictive maintenance strategy.

[0128] Exemplarily, first obtain the microcrack growth rate data through a crack growth rate monitoring device. After this data is processed by the mapping of fracture mechanics theory and then subjected to time integral operation, the real-time crack length value can be obtained. Taking a certain steel key bearing as an example, when the detected crack growth rate is 0.02 mm per cycle, after 1000 cycles of loading, the calculated real-time crack length is 20 mm. Combining with the material fracture toughness value (the fracture toughness of this steel is 120 MPa·m 0.5 ) to calculate the initial micro-residual strength, and then multiplying it by the crack closure effect correction coefficient (taking the value of 0.85) to obtain the corrected micro-residual strength value.

[0129] Meanwhile, a fatigue damage detection system is used to obtain data on the macroscopic fatigue damage degree. This data is input into the non-linear cumulative damage theory model and corrected considering the influence of load interaction. Taking a certain test condition as an example, the original damage amount is 0.45, and the damage amount after correction by load interaction is 0.52. Substituting the corrected damage amount into the strength degradation equation, the calculated strength degradation rate is 0.15% per hour. Through time integration operation, the macroscopic strength value of the current component can be obtained.

[0130] A micro-macro dual-scale damage coupling relationship model is established. The corrected micro residual strength and the current macroscopic strength data are input into the coupling equation to calculate the dynamic correlation coefficient. This coefficient reflects the degree of mutual influence between micro crack propagation and macroscopic fatigue damage. Substituting the dynamic correlation coefficient into the comprehensive failure criterion equation, the failure criterion value is obtained.

[0131] Dividing the calculated failure criterion value by the allowable stress of the material (taking the value of 250 MPa), the failure criterion ratio is obtained. According to the pre-established risk level classification standard, when the ratio is in the range of 0 - 0.5, it is a low risk, in the range of 0.5 - 0.8 it is a medium risk, and above 0.8 it is a high risk. The failure probability is calculated through the failure distribution function and weighted with the risk level to finally obtain the evaluation result of the component health state.

[0132] The operating parameters of the equipment are collected in real time, including data such as load, operating time, and environmental parameters. Taking a certain rotating equipment as an example, the rated load collected is 80%, the continuous operating time reaches 5000 hours, and the environmental temperature is 35 degrees Celsius. The component health state evaluation result, the operating parameter data, and the maintenance resource constraint conditions are input into the maintenance optimization model to solve for the optimal maintenance time interval.

[0133] According to the calculated optimal maintenance time interval and the equipment operating condition parameters, combined with the predictive maintenance strategy generation model, an initial maintenance plan is formulated. Continuously collect the real-time condition data of the equipment. When it is detected that the data exceeds the preset range, the latest data is promptly input into the maintenance optimization model for recalculation to dynamically update the maintenance strategy.

[0134] In this embodiment, by establishing the coupling relationship between micro crack propagation and macroscopic fatigue damage, the accurate characterization of the multi-scale damage evolution process is realized, and the accuracy of failure risk assessment is improved. Using the dynamic correlation coefficient and the comprehensive failure criterion, a scientific risk level classification standard is established, making the evaluation result of the equipment health state more objective and reliable. Based on the dynamic optimization method of real-time condition data, the adaptive adjustment of the maintenance strategy is realized, which not only ensures the operating safety of the equipment, but also improves the maintenance efficiency and reduces the maintenance cost.

[0135] In an alternative embodiment, substituting the optimal maintenance time interval and the equipment operating condition parameters into a pre-constructed predictive maintenance strategy generation model, the obtained initial predictive maintenance strategy includes:

[0136] Dividing the optimal maintenance time interval sequence into multiple time sub-intervals according to its change rate to obtain a time sub-interval sequence; calculating the mutual information and conditional entropy between the equipment operating condition parameters, and determining a set of key condition parameters based on the mutual information and conditional entropy;

[0137] For the time sub-interval sequence, calculating the state embedding features of each time sub-interval and performing weighted fusion to obtain a time feature vector; constructing a parameter correlation graph based on the set of key condition parameters, and extracting the condition feature vector of the parameter correlation graph;

[0138] Inputting the time feature vector and the condition feature vector into a pre-constructed hierarchical strategy generation network; wherein, the first layer generates a maintenance timing template based on the time feature vector, the second layer generates a maintenance action configuration based on the condition feature vector, and the third layer fuses the maintenance timing template and the maintenance action configuration to generate a candidate maintenance strategy;

[0139] Collecting the performance parameters of the equipment before and after maintenance, calculating the recovery degree of the performance parameters after maintenance relative to the performance parameters before maintenance to obtain a maintenance effect index; monitoring the degradation state of the key components of the equipment, and calculating the failure risk probability based on the degradation rate and the remaining life prediction result to obtain a reliability index; counting the labor cost, spare parts cost and downtime loss during the maintenance process, and calculating the total maintenance cost per unit time to obtain a cost index;

[0140] Optimally adjusting the parameters of the hierarchical strategy generation network according to the maintenance effect index, the reliability index and the cost index. When the maintenance effect index is greater than a first preset threshold, the reliability index is greater than a second preset threshold, and the cost index is less than a third preset threshold, the current candidate maintenance strategy is determined as the initial predictive maintenance strategy.

[0141] Exemplarily, first, the system receives the optimal maintenance time interval sequence and the equipment operating condition parameter data. For the time interval sequence, it is divided by calculating the change rate between adjacent time points. For example, when the change rate exceeds a preset value (such as 20%), it is used as the starting point of a new time sub-interval. For the condition parameters, the system calculates the mutual information and conditional entropy between the parameters. Specifically, for parameters such as temperature, pressure, and vibration, when the mutual information value is greater than 0.8 and the conditional entropy is less than 0.2, they are included in the set of key condition parameters.

[0142] In terms of time feature extraction, a long short-term memory network is used to process the data of each time sub-interval. The network contains 64 hidden units, the input is time series data, and the output is a 128-dimensional state embedding vector. Subsequently, an attention mechanism is used to weight the state embeddings of different time periods. The weight coefficients are normalized by the softmax function to obtain the final 256-dimensional time feature vector.

[0143] For working condition feature extraction, the system constructs a parameter correlation graph based on key working condition parameters. The nodes in the graph represent parameters, and the edges represent the correlation strength between parameters. A graph convolutional network is used for feature extraction, which contains 3 graph convolutional layers, with 32, 64, and 128 convolutional kernels in each layer respectively. After self-attention processing, a 192-dimensional working condition feature vector is obtained.

[0144] The hierarchical strategy generation network adopts a three-layer architecture. The first layer uses a fully connected network, the input is the time feature vector, and the output maintains a timing template, including suggestions for maintenance time points. The second layer generates specific maintenance action configurations based on the working condition feature vector, such as combinations of operations like replacement, inspection, cleaning, etc. The third layer fuses the timing template and the action configuration to generate a complete maintenance strategy.

[0145] In the process of policy optimization, the value network evaluates the state-action value of candidate policies. The value network contains 3 fully connected layers, and the number of hidden layer units is 256, 128, and 64 respectively. According to the evaluation results, the parameter update of the policy network is guided, the learning rate is set to 0.001, and the discount factor is 0.95.

[0146] The system collects maintenance effect data, including the equipment performance parameters before and after maintenance. For example, if the vibration value of a certain bearing was 5 mm / s before maintenance and dropped to 2 mm / s after maintenance, the degree of performance recovery is 60%. By monitoring the degradation state of key components, the remaining life is predicted and the failure risk is calculated. At the same time, the maintenance cost is counted, including labor costs, spare parts costs, and downtime losses, etc.

[0147] When the maintenance effect index exceeds 85%, the reliability index exceeds 90%, and the maintenance cost per unit time is lower than 80% of the budget target, the system determines the current candidate policy as the initial predictive maintenance policy.

[0148] Figure 3 This is a performance comparison graph of the predictive maintenance strategy generation model in the embodiments of the present invention, as Figure 3As shown, this figure comprehensively compares the performance of three methods in four key evaluation indicators. The horizontal axis represents the four evaluation indicators of maintenance effect, reliability, cost - effectiveness, and comprehensive score, and the vertical axis represents the percentage performance score (0 - 100%). In the figure, three groups of bar charts with different styles are used to represent "Prior Art (RUL - based CBM)", "Traditional Method (threshold - based maintenance)", and "This Method (Hierarchical Strategy Generation Network)". The data shows that in terms of maintenance effect, the scores of the three methods are 68%, 79%, and 85% respectively; in terms of reliability, the scores are 56%, 72%, and 83% respectively; in terms of cost - effectiveness, the scores are 75%, 64%, and 79% respectively; in terms of comprehensive score, the three methods obtain scores of 66.3%, 71.7%, and 82.3% respectively. The red dashed line in the figure marks the critical threshold line of 75%. It can be seen that this method exceeds the critical threshold in all indicators, while the prior art is insufficient in terms of reliability and maintenance effect, and the traditional method is below the threshold in terms of cost - effectiveness. This comparison fully demonstrates the superiority of the hierarchical strategy generation network in balancing maintenance effect, reliability, and cost - effectiveness, providing intuitive data support for enterprises to select appropriate predictive maintenance strategies.

[0149] In this embodiment, by fusing time characteristics and working condition characteristics to generate a maintenance strategy, the timing and state correlation of equipment operation are fully considered, improving the pertinence and adaptability of the maintenance strategy. Using a hierarchical strategy generation network structure, the maintenance decision is decomposed into two levels: timing arrangement and action configuration, which not only ensures the integrity of the strategy but also improves the accuracy of each link. By introducing multi - dimensional evaluation indicators and an iterative optimization mechanism, while ensuring the maintenance effect and equipment reliability, the maintenance cost is effectively controlled, realizing the dynamic optimization and continuous improvement of the maintenance strategy.

[0150] In an alternative embodiment, the edge computing unit receives and parses the adaptive compensation control instruction, establishes an active compensation control model based on intelligent materials, and calculates the optimal ratio of each compensation force and controls the corresponding actuators, including:

[0151] Extract the compensation target displacement value, compensation time threshold, and compensation accuracy threshold from the adaptive compensation control instruction;

[0152] Collect the data of the environmental temperature sensor to obtain the temperature change value, calculate the thermal expansion displacement value of the piezoelectric material and the thermal expansion displacement value of the magnetostrictive material respectively according to the temperature change value, establish a piezoelectric material displacement response equation, substitute the piezoelectric constant, piezoelectric drive voltage, and the thermal expansion displacement value of the piezoelectric material into the piezoelectric material displacement response equation to obtain the piezoelectric material compensation displacement value; at the same time, calculate the magnetostrictive material compensation displacement value according to the magnetostrictive coefficient, magnetostrictive drive magnetic field intensity, and the thermal expansion displacement value of the magnetostrictive material;

[0153] Establish an intelligent material collaborative compensation mechanism including the piezoelectric material ratio coefficient and the magnetostrictive material ratio coefficient. Multiply the piezoelectric material compensation displacement value by the piezoelectric material ratio coefficient to obtain the piezoelectric material weighted displacement value, and multiply the magnetostrictive material compensation displacement value by the magnetostrictive material ratio coefficient to obtain the magnetostrictive material weighted displacement value;

[0154] Based on the intelligent material collaborative compensation mechanism, add the piezoelectric material weighted displacement value and the magnetostrictive material weighted displacement value to obtain the total compensation displacement value; perform non-linear optimization to solve the total compensation displacement value, the compensation target displacement value, the compensation time threshold, and the compensation accuracy threshold to obtain the optimal piezoelectric material ratio coefficient and the optimal magnetostrictive material ratio coefficient;

[0155] Substitute the optimal piezoelectric material ratio coefficient, the compensation target displacement value, and the piezoelectric material thermal expansion displacement value into the piezoelectric material displacement response equation to inversely solve for the target piezoelectric drive voltage. At the same time, substitute the optimal magnetostrictive material ratio coefficient, the compensation target displacement value, and the magnetostrictive material thermal expansion displacement value into the magnetostrictive material displacement response equation to inversely solve for the target magnetic field drive intensity;

[0156] Collect displacement sensor data to obtain the real-time compensation displacement value, and compare the real-time compensation displacement value with the compensation target displacement value to obtain the displacement tracking error value;

[0157] Based on the displacement tracking error value, adjust the target piezoelectric drive voltage and the target magnetic field drive intensity in real time to obtain the compensation drive voltage and the compensation magnetic field intensity; control the piezoelectric material actuator and the magnetostrictive material actuator according to the compensation drive voltage and the compensation magnetic field intensity to achieve the adaptive active compensation of the weighing mechanism.

[0158] Exemplarily, after receiving the adaptive compensation control instruction, it is necessary to extract the compensation target displacement value, the compensation time threshold, and the compensation accuracy threshold. The compensation target displacement value refers to the displacement adjustment target that the weighing mechanism needs to achieve during the compensation process; the compensation time threshold represents the time limit required to complete the compensation operation; the compensation accuracy threshold defines the error tolerance range of the compensation result. These parameters serve as the basic information for subsequent compensation control and run through the entire compensation control process.

[0159] Collect the data of the environmental temperature sensor to obtain the temperature change value in real time. The temperature change value represents the change amount of the external environmental temperature relative to the initial state, which is an important basis for evaluating the thermal expansion effect of intelligent materials. Calculate the thermal expansion displacement values of piezoelectric materials and magnetostrictive materials based on this value. The thermal expansion displacement value refers to the dimensional change of the material caused by temperature change, which can be used to measure the degree to which the material performance is affected by the environmental temperature. For example, when the temperature rises by 10 °C, a certain piezoelectric material may generate a thermal expansion displacement of about 0.02 mm. By combining the piezoelectric constant (a parameter describing the electro-mechanical coupling performance of piezoelectric materials), the piezoelectric driving voltage (the voltage signal applied to the piezoelectric material), and the temperature change value, construct the displacement response equation of the piezoelectric material, and further calculate its compensation displacement value.

[0160] Establish a cooperative compensation mechanism for intelligent materials to comprehensively compensate the effects of piezoelectric materials and magnetostrictive materials. The cooperative compensation mechanism includes the piezoelectric material ratio coefficient and the magnetostrictive material ratio coefficient, which are used to describe the contribution ratio of the two materials to the total compensation displacement. For example, when the ratio coefficients are set to 0.6 and 0.4 respectively, it means that the piezoelectric material contributes 60% of the compensation displacement, while the magnetostrictive material contributes 40%. Multiply the compensation displacement value of the piezoelectric material by its ratio coefficient to obtain the weighted displacement value of the piezoelectric material; the magnetostrictive material calculates its weighted displacement value in the same way. After adding the two weighted displacement values, the total compensation displacement value is generated.

[0161] Input the total compensation displacement value, the compensation target displacement value, the compensation time threshold, and the compensation accuracy threshold into the non-linear optimization solver to optimize and calculate the optimal piezoelectric material ratio coefficient and magnetostrictive material ratio coefficient. For example, the initially calculated total compensation displacement is 0.5 mm, but the target displacement is 0.6 mm. The solver adjusts the ratio coefficients to obtain the optimized values to ensure that the final compensation reaches the target.

[0162] Using the optimized ratio coefficients, substitute the compensation target displacement value and the thermal expansion displacement value of the piezoelectric material into the displacement response equation of the piezoelectric material to inversely solve the target piezoelectric driving voltage. Similarly, substitute the compensation target displacement value and the thermal expansion displacement value of the magnetostrictive material into the displacement response equation of the magnetostrictive material to inversely solve the target magnetic field driving intensity. The target piezoelectric driving voltage represents the voltage signal that needs to be applied to the piezoelectric material, and the target magnetic field driving intensity represents the magnetic field intensity that needs to be applied to the magnetostrictive material. Both are important inputs for precisely controlling the compensation behavior of the material.

[0163] Collect the data of the displacement sensor in real time to obtain the real-time compensation displacement value during the compensation process. The real-time compensation displacement value refers to the immediate displacement adjustment achieved by the intelligent material during actual operation. Compare this value with the compensation target displacement value to calculate the displacement tracking error value. The displacement tracking error value reflects the deviation between the current compensation state and the target state, which is an important basis for guiding subsequent control adjustments.

[0164] An active compensation controller is established based on the displacement tracking error value. The active compensation controller dynamically adjusts the target piezoelectric drive voltage and the target magnetic field drive intensity through real-time calculation. For example, when the displacement tracking error value is 0.01 mm, the controller can increase the piezoelectric drive voltage or adjust the magnetic field intensity to reduce the error. The controller parameters are tuned using the critical oscillation method to ensure a balance between the response speed and stability of the system. Finally, the compensated drive voltage and the compensated magnetic field intensity are output by the controller, and these signals directly act on the piezoelectric material actuator and the magnetostrictive material actuator to complete the compensation operation.

[0165] After the compensation is completed, relevant data is collected to construct an evaluation index system for the compensation effect. The evaluation indexes include the accuracy of the compensated displacement (whether the deviation between the actual displacement and the target displacement is within the allowable range), the response time (whether the time required to complete the compensation meets the threshold requirement), and the system stability (whether the performance remains consistent during multiple compensations). After generating the compensation effect evaluation report, it is possible to analyze whether the system performance meets the design objectives and provide a basis for subsequent optimization.

[0166] In this embodiment, by combining multiple intelligent materials, a collaborative compensation mechanism is constructed to effectively improve the adaptive ability of the weighing mechanism in a complex environment and significantly enhance the precise compensation effect of the system. Utilizing the characteristics of piezoelectric materials and magnetostrictive materials, it is possible to respond in real time to the influence of environmental temperature changes on the weighing mechanism, reduce the displacement error caused by thermal expansion, and ensure that the weighing accuracy is always at a high level. Through non-linear optimization and solution, the mixing ratio coefficients of the compensation materials are dynamically adjusted, enabling a more flexible and efficient compensation process, reducing energy consumption and extending the service life of the equipment. By real-time collecting and analyzing the compensated displacement data and establishing a feedback control mechanism, the displacement tracking error can be quickly responded to, further improving the stability and control accuracy of the system. In addition, the introduction of the compensation effect evaluation index system not only provides a clear performance measurement standard but also provides a scientific basis for system optimization and improvement, forming a complete closed-loop from compensation to evaluation and then to optimization.

[0167] Figure 4 The following is a schematic structural diagram of a remote maintenance and intelligent inspection and warning management system for a weight measuring instrument according to an embodiment of the present invention, as Figure 4 shown, the system includes:

[0168] A first unit for setting a multi-source sensor array on the weighing mechanism of the weight measuring instrument to collect detection data, performing a three-domain joint analysis in the time domain, frequency domain, and wavelet domain, extracting multi-dimensional coupling features, generating an instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism through a deep fusion model, calculating the working state features of the weighing mechanism, forming a multi-dimensional health feature vector, and transmitting it to a remote monitoring platform;

[0169] The second unit is used for the remote monitoring platform to receive the multi-dimensional health feature vectors and input them into a pre-established multi-scale damage evolution prediction model. Among them, the microscopic prediction sub-module calculates the microscopic crack growth rate based on fracture mechanics, and the macroscopic prediction sub-module analyzes the macroscopic fatigue damage degree based on the cumulative damage theory. According to the coupling relationship between the microscopic crack growth rate and the macroscopic fatigue damage degree, the failure risk level of key components is determined, a predictive maintenance strategy is generated, and an adaptive compensation control instruction is generated and sent to the edge computing unit;

[0170] The third unit is used for the edge computing unit to receive and parse the adaptive compensation control instruction, establish an active compensation control model based on intelligent materials, calculate the optimal ratio of each compensation force and control the corresponding actuator, collect the real-time response data during the compensation process, construct an evaluation index system for the compensation effect to generate a compensation effect evaluation report, online optimize the parameters of the multi-scale damage evolution prediction model based on the compensation effect evaluation report, and at the same time update the predictive maintenance strategy library to form a closed-loop control of prediction-compensation-evaluation-optimization.

[0171] In the third aspect of the embodiments of the present invention,

[0172] A kind of electronic device is provided, including:

[0173] A processor;

[0174] A memory for storing instructions executable by the processor;

[0175] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0176] In the fourth aspect of the embodiments of the present invention,

[0177] A computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0178] The present invention can be a method, device, system and / or computer program product. The computer program product can include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are uploaded.

[0179] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote maintenance and intelligent inspection and warning management method for a weight measuring instrument, characterized in that, Including: A multi-source sensor array is set on the weighing mechanism of the weight measuring instrument to collect detection data, and three-domain joint analysis in the time domain, frequency domain, and wavelet domain is carried out to extract multi-dimensional coupling features. Through a deep fusion model, an instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism are generated, the working state characteristics of the weighing mechanism are calculated, a multi-dimensional health feature vector is formed, and it is transmitted to the remote monitoring platform; The remote monitoring platform receives the multi-dimensional health feature vector and inputs it into a pre-established multi-scale damage evolution prediction model. Among them, the micro prediction sub-module calculates the micro crack propagation rate based on fracture mechanics, and the macro prediction sub-module analyzes the macro fatigue damage degree based on the cumulative damage theory. According to the coupling relationship between the micro crack propagation rate and the macro fatigue damage degree, the failure risk level of key components is determined, a predictive maintenance strategy is generated, and an adaptive compensation control instruction is generated and sent to the edge computing unit; The edge computing unit receives and analyzes the adaptive compensation control instruction, establishes an active compensation control model based on intelligent materials, calculates the optimal ratio of each compensation force and controls the corresponding actuator, collects real-time response data during the compensation process, constructs an evaluation index system for the compensation effect to generate a compensation effect evaluation report, online optimizes the parameters of the multi-scale damage evolution prediction model based on the compensation effect evaluation report, and at the same time updates the predictive maintenance strategy library to form a closed-loop control of prediction-compensation-evaluation-optimization.

2. The method according to claim 1, characterized in that A multi-source sensor array is set on the weighing mechanism of the weight measuring instrument to collect detection data, and three-domain joint analysis in the time domain, frequency domain, and wavelet domain is carried out to extract multi-dimensional coupling features. Generating an instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism includes: The wireless ad hoc network sensing network is used to transmit the detection data collected by the multi-source sensors to the edge computing unit built in the weighing mechanism. The master-slave clock synchronization mechanism is used to align the time of the detection data, and the sampling frequency difference is compensated through the data cache queue with adaptive jitter compensation to obtain time-synchronized detection data; The wavelet basis function is used to denoise the detection data. For the temperature data in the denoised detection data, the temperature drift of the strain data is eliminated to obtain the compensated strain data; The system natural frequency is extracted from the vibration data in the denoised detection data through the fast Fourier transform, the modal damping ratio is estimated by the half-power bandwidth method, and the modulation frequency characteristics are extracted through the Hilbert transform and envelope spectrum analysis to form a frequency domain feature vector; The continuous wavelet transform is performed on the acceleration data to obtain the time-frequency distribution characteristics, the energy distribution of the wavelet coefficients of the time-frequency distribution characteristics is calculated to obtain the wavelet energy characteristics, and the multi-dimensional coupling characteristics are extracted by performing correlation analysis on the wavelet energy characteristics and the frequency domain feature vector; A multi-layer convolutional structure with residual connections is used to encode the multi-dimensional coupled features. Through skip connections, multi-scale features are retained and mapped to the target space to obtain target features. A multi-head self-attention mechanism is used to adaptively weight and fuse the target features to obtain fused features. Through deconvolution operations, the fused features and the compensated strain data are reconstructed to generate the instantaneous dynamic load distribution map and stress field distribution map of the weighing mechanism.

3. The method according to claim 1, wherein Calculate the working state features of the weighing mechanism to form a multi-dimensional health feature vector, including: Build a deep fusion model containing a convolutional neural network and a recurrent neural network, obtain the measurement data of the weighing mechanism, input the measurement data into the deep fusion model, and generate the instantaneous dynamic load distribution map and stress field distribution map of the weighing mechanism as the data to be processed; Based on the data to be processed, perform adaptive mesh division on the weighing mechanism, increase the local mesh density in the area where the stress concentration is greater than the preset concentration threshold to obtain mesh elements, and calculate the local strain energy density of the mesh elements. The strain energy distribution characteristics of the overall structure are obtained by accumulating the strain energy densities of each mesh element; Adopt a finite element stress reconstruction method based on the stress function to construct a biharmonic stress function, establish a stress component mapping relationship, interpolate the strain data of discrete measurement points to obtain weight coefficients, and reconstruct the stress field distribution map in combination with the strain energy distribution characteristics; Combine the reconstructed stress field distribution map with the instantaneous dynamic load distribution map, extract the local stress concentration coefficient, modal vibration intensity index, load distribution non-uniformity index, and cumulative damage degree index, and select the principal components with a preset cumulative contribution rate through principal component analysis to construct an initial multi-dimensional health feature vector; Use a long short-term memory network to online update the initial multi-dimensional health feature vector to obtain a state vector, and use an attention mechanism to weight and fuse the state vectors at different time scales to obtain the final multi-dimensional health feature vector.

4. The method according to claim 1, wherein The remote monitoring platform receives the multi-dimensional health feature vector and inputs it into a pre-established multi-scale damage evolution prediction model. Among them, the microscopic prediction sub-module calculates the microscopic crack propagation rate based on fracture mechanics, and the macroscopic prediction sub-module analyzes the macroscopic fatigue damage degree based on the cumulative damage theory, including: Use wavelet transform to perform multi-scale decomposition on the multi-dimensional health feature vector to obtain high-frequency component signals and low-frequency component signals; Input the high-frequency component signal into the microscopic prediction sub-module, obtain the time-frequency energy spectrum through wavelet transform, extract the change characteristics of the stress intensity factor, establish the mapping relationship of the stress field at the crack tip, and obtain the stress field distribution at the crack tip; Based on the stress field distribution at the crack tip, divide the crack area into finite element meshes and perform adaptive encryption to construct a node network including the topological relationship of mesh nodes; use the stress field distribution at the crack tip, material constitutive parameters, crack propagation direction, and critical stress intensity factor as the node features of the node network; use a graph convolutional neural network and an attention mechanism to extract dynamic spatial correlation features; Calculate the critical crack propagation length according to the dynamic spatial correlation characteristics; establish the functional relationship between the propagation rate and the stress intensity factor by using the crack propagation rate formula considering the crack closure effect and the overload effect, and obtain the microscopic crack propagation rate considering the crack closure effect and the overload effect. Input the low-frequency component signal into the macroscopic prediction sub-module to obtain the stress amplitude spectrum and the cyclic number distribution characteristics. Input the cyclic number distribution characteristics into the long short-term memory network to calculate the damage feature weights, and obtain the time-series dependence features through the time-scale attention mechanism. Input the damage feature weights and the time-series dependence features into the analysis unit based on the cumulative damage theory, and calculate the overall fatigue state of the weighing mechanism by using the non-linear cumulative damage theory. Combine the overall fatigue state with the fatigue life curve, and calculate the equivalent stress amplitude through the damage equivalence criterion to obtain the macroscopic fatigue damage degree.

5. The method according to claim 1, characterized in that, Determine the failure risk level of the key components according to the coupling relationship between the microscopic crack propagation rate and the macroscopic fatigue damage degree, and generate the predictive maintenance strategy including: Input the microscopic crack propagation rate into the mapping relationship of fracture mechanics theory, and perform integral calculation on the microscopic crack propagation rate to obtain the real-time crack length. Calculate the initial microscopic remaining strength based on the real-time crack length and the material fracture toughness. Multiply the initial microscopic remaining strength by the crack closure effect correction coefficient to obtain the corrected microscopic remaining strength. Perform load interaction correction on the macroscopic fatigue damage degree to obtain the corrected damage amount. Calculate the strength degradation rate according to the corrected damage amount. Perform time integration on the strength degradation rate to obtain the current macroscopic strength. Perform two-scale damage coupling calculation on the corrected microscopic remaining strength and the current macroscopic strength to obtain the dynamic correlation coefficient. Establish the comprehensive failure criterion equation, and substitute the dynamic correlation coefficient into the comprehensive failure criterion equation to calculate the comprehensive failure criterion value. Divide the comprehensive failure criterion value by the allowable stress of the material to obtain the failure criterion ratio. Establish the risk level classification standard, and compare the failure criterion ratio with the risk level classification standard to obtain the failure risk level. Establish the failure distribution function, substitute the failure risk level into the failure distribution function to calculate the failure probability, and perform weighted calculation on the failure risk level and the failure probability to obtain the component health state evaluation result. Collect the equipment operating condition parameters to establish the maintenance optimization model, and substitute the component health state evaluation result and the equipment operating condition parameters into the maintenance optimization model for solution to obtain the optimal maintenance time interval. Substitute the optimal maintenance time interval and the equipment operating condition parameters into the pre-constructed predictive maintenance strategy generation model to obtain the initial predictive maintenance strategy. Collect the real-time equipment condition data in real time, compare the real-time equipment condition data with the preset condition range, and when the real-time equipment condition data exceeds the preset condition range, substitute the real-time equipment condition data into the maintenance optimization model to recalculate and update the predictive maintenance strategy.

6. The method according to claim 5, wherein Substitute the optimal maintenance time interval and the equipment operating condition parameters into the pre-constructed predictive maintenance strategy generation model to obtain the initial predictive maintenance strategy including: Divide the optimal maintenance time interval sequence into multiple time sub - intervals according to its change rate to obtain a time sub - interval sequence; calculate the mutual information and conditional entropy between the operating condition parameters of the device, and determine the set of key operating condition parameters based on the mutual information and conditional entropy; For the time sub - interval sequence, calculate the state embedding features of each time sub - interval and perform weighted fusion to obtain a time feature vector; construct a parameter correlation graph based on the set of key operating condition parameters, and extract the operating condition feature vector of the parameter correlation graph; Input the time feature vector and the operating condition feature vector into a pre - constructed hierarchical policy generation network; among them, the first layer generates a maintenance time sequence template based on the time feature vector, the second layer generates a maintenance action configuration based on the operating condition feature vector, and the third layer fuses the maintenance time sequence template and the maintenance action configuration to generate a candidate maintenance policy; Collect the performance parameters of the device before and after maintenance, calculate the recovery degree of the performance parameters after maintenance relative to the performance parameters before maintenance to obtain a maintenance effect index; monitor the degradation state of the key components of the device, and calculate the failure risk probability based on the degradation rate and the remaining life prediction result to obtain a reliability index; count the labor cost, spare part cost and downtime loss during the maintenance process, and calculate the total maintenance cost per unit time to obtain a cost index; Optimize and adjust the parameters of the hierarchical policy generation network according to the maintenance effect index, reliability index and cost index. When the maintenance effect index is greater than the first preset threshold, the reliability index is greater than the second preset threshold, and the cost index is less than the third preset threshold, determine the current candidate maintenance policy as the initial predictive maintenance policy.

7. The method according to claim 1, wherein The edge computing unit receives and parses the adaptive compensation control instruction, establishes an active compensation control model based on intelligent materials, and calculates the optimal ratio of each compensation force and controls the corresponding actuator, including: Extract the compensation target displacement value, compensation time threshold and compensation accuracy threshold from the adaptive compensation control instruction; Collect the data of the environmental temperature sensor to obtain the temperature change value, calculate the thermal expansion displacement value of the piezoelectric material and the thermal expansion displacement value of the magnetostrictive material respectively according to the temperature change value, establish a piezoelectric material displacement response equation, and substitute the piezoelectric constant, piezoelectric driving voltage and the thermal expansion displacement value of the piezoelectric material into the piezoelectric material displacement response equation to obtain the piezoelectric material compensation displacement value; at the same time, calculate the magnetostrictive material compensation displacement value according to the magnetostrictive coefficient, magnetostrictive driving magnetic field intensity and the thermal expansion displacement value of the magnetostrictive material; Establish an intelligent material collaborative compensation mechanism including the piezoelectric material ratio coefficient and the magnetostrictive material ratio coefficient, multiply the piezoelectric material compensation displacement value by the piezoelectric material ratio coefficient to obtain the piezoelectric material weighted displacement value, and multiply the magnetostrictive material compensation displacement value by the magnetostrictive material ratio coefficient to obtain the magnetostrictive material weighted displacement value; Based on the intelligent material collaborative compensation mechanism, add the piezoelectric material weighted displacement value and the magnetostrictive material weighted displacement value to obtain the total compensation displacement value; perform non - linear optimization on the total compensation displacement value, compensation target displacement value, compensation time threshold and compensation accuracy threshold to obtain the optimal piezoelectric material ratio coefficient and the optimal magnetostrictive material ratio coefficient; Substitute the optimal piezoelectric material ratio coefficient, the compensated target displacement value, and the thermal expansion displacement value of the piezoelectric material into the piezoelectric material displacement response equation, and inversely solve to obtain the target piezoelectric driving voltage. At the same time, substitute the optimal magnetostrictive material ratio coefficient, the compensated target displacement value, and the thermal expansion displacement value of the magnetostrictive material into the magnetostrictive material displacement response equation, and inversely solve to obtain the target magnetic field driving intensity; Collect the data of the displacement sensor to obtain the real-time compensated displacement value, and compare the real-time compensated displacement value with the compensated target displacement value to obtain the displacement tracking error value; Based on the displacement tracking error value, adjust the target piezoelectric driving voltage and the target magnetic field driving intensity in real time to obtain the compensated driving voltage and the compensated magnetic field intensity; control the piezoelectric material actuator and the magnetostrictive material actuator according to the compensated driving voltage and the compensated magnetic field intensity to achieve the adaptive active compensation of the weighing mechanism.

8. A remote maintenance and intelligent inspection and warning management system for a weight measuring instrument, which is used to implement the method described in any one of the foregoing claims 1-7, and is characterized in that, Comprising: The first unit is configured to set a multi-source sensor array on the weighing mechanism of the weight meter to collect detection data, perform a three-domain joint analysis of the time domain - frequency domain - wavelet domain, extract multi-dimensional coupling features, generate an instantaneous dynamic load distribution map and a stress field distribution map of the weighing mechanism through a deep fusion model, calculate the working state features of the weighing mechanism, form a multi-dimensional health feature vector, and transmit it to the remote monitoring platform; The second unit is configured to receive the multi-dimensional health feature vector by the remote monitoring platform and input it into a pre-established multi-scale damage evolution prediction model. Among them, the microscopic prediction sub-module calculates the microscopic crack propagation rate based on fracture mechanics, the macroscopic prediction sub-module analyzes the macroscopic fatigue damage degree based on the cumulative damage theory, determines the failure risk level of key components according to the coupling relationship between the microscopic crack propagation rate and the macroscopic fatigue damage degree, generates a predictive maintenance strategy, and generates an adaptive compensation control instruction, and sends it to the edge computing unit; The third unit is configured to receive and parse the adaptive compensation control instruction by the edge computing unit, establish an active compensation control model based on intelligent materials, calculate the optimal ratio of each compensation force and control the corresponding actuator, collect the real-time response data during the compensation process, construct an evaluation index system for the compensation effect to generate a compensation effect evaluation report, online optimize the parameters of the multi-scale damage evolution prediction model based on the compensation effect evaluation report, and at the same time update the predictive maintenance strategy library to form a closed-loop control of prediction - compensation - evaluation - optimization.

9. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images