Weighing sensor failure time prediction method, device, equipment and storage medium

Abstract

The present application relates to the field of time prediction technology and discloses a method, apparatus, device, and storage medium for predicting the fault time of a weighing sensor, including: inputting the working status data of the current weighing sensor into a fault detection model to obtain candidate fault features; obtaining load fluctuation data through the historical load curve of the current weighing sensor, and preprocessing the load fluctuation data and candidate fault features to obtain load features and fault features; extracting time series features from the load features and fault features based on a hybrid convolutional neural network to obtain load time series features and fault time series features; inputting the load time series features and fault time series features into a fault time prediction model to predict the fault time of the current weighing sensor. The hybrid convolutional network is used to capture the long-term and short-term correlation between load and fault, avoiding reliance on a single feature, improving the real-time and accuracy of the prediction, and achieving end-to-end fault time prediction, which is suitable for high-volatility industrial scenarios.

Description

Technical Field

[0001] The present application relates to the field of time prediction technology, and in particular to a method, device, equipment and storage medium for predicting the failure time of a weighing sensor. Background Art

[0002] Load cells, core components of truck scales, operate in complex and ever-changing environments, with their performance affected by numerous factors such as temperature, humidity, and tilt angle. The lack of effective lifecycle prediction methods makes it difficult for companies to accurately assess their service life, hindering the development of targeted maintenance and replacement plans. Summary of the Invention

[0003] The main purpose of this application is to provide a weighing sensor failure time prediction method, device, equipment and storage medium, aiming to solve the technical problem that the existing technology lacks effective weighing sensor life cycle prediction means, which makes it impossible for enterprises to formulate targeted maintenance and replacement plans in advance.

[0004] To achieve the above objectives, the present application proposes a method for predicting the failure time of a weighing sensor, the method comprising:

[0005] Input the working status data of the current weighing sensor into the fault detection model to obtain candidate fault features;

[0006] Acquire load fluctuation data through a historical load curve of the current load cell, and pre-process the load fluctuation data and the candidate fault features to obtain a load feature and a fault feature;

[0007] Performing time series feature extraction on the load feature and the fault feature based on a hybrid convolutional neural network to obtain a load time series feature and a fault time series feature;

[0008] The load time series characteristics and the fault time series characteristics are input into a fault time prediction model to predict the fault time of the current weighing sensor.

[0009] Optionally, the step of inputting the current working state data of the weighing sensor into the fault detection model to obtain candidate fault features includes:

[0010] Processing the working state data of the current weighing sensor through a sliding average filter template to obtain first state data;

[0011] removing the first state data from the working state data to obtain second state data;

[0012] The second state data and the working state data are weightedly fused to obtain third state data;

[0013] The third state data is input into a preset fault detection model to obtain candidate fault features.

[0014] Optionally, the step of inputting the third state data into a preset fault detection model to obtain candidate fault features includes:

[0015] determining a candidate fault interval of the third state data by an improved density peak search method;

[0016] Performing local abnormality discrimination on the candidate fault interval to generate a fault probability distribution map;

[0017] performing noise suppression on the fault probability distribution graph by an adaptive filtering method to obtain a suppressed fault interval, and determining a target fault interval corresponding to the second state data based on the suppressed fault interval;

[0018] Dynamic threshold segmentation is performed on the target fault interval to obtain candidate fault features.

[0019] Optionally, the step of determining the candidate fault interval of the third state data by using an improved density peak search method includes:

[0020] Extracting the maximum overload value and the number of overload times at each time point in the third state data;

[0021] Calculating the load density at the time point based on the number of overloads at adjacent time points;

[0022] Determine a load density peak point based on the maximum overload value and the load density by combining weights;

[0023] A time window to be detected is constructed with the load density peak point as the center, an overload probability value within the time window to be detected is calculated, and the overload probability values ​​are sorted in descending order to obtain the candidate fault interval.

[0024] Optionally, the step of performing local abnormality identification on the candidate fault interval and generating a fault probability distribution map includes:

[0025] Obtaining a first load mean and a first gradient mean of a current time window in the candidate fault interval, and a second load mean and a second gradient mean of an adjacent window;

[0026] determining a load deviation according to the first load mean and the second load mean, and determining a gradient mean difference according to the first gradient mean and the second gradient mean;

[0027] Determining a directional consistency factor based on the positive and negative signs of the gradient mean difference;

[0028] If the directional consistency factor satisfies the multi-directional fluctuation condition, marking the candidate fault interval as a normal interval;

[0029] If the directional consistency factor does not meet the multi-directional fluctuation condition, a fault probability distribution diagram is generated according to the load deviation, the gradient mean difference and the directional consistency factor.

[0030] Optionally, the step of performing time series feature extraction on the load feature and the fault feature based on a hybrid convolutional neural network to obtain load time series features and fault time series features includes:

[0031] Inputting the load feature and the fault feature into the convolution layer of the hybrid convolutional neural network respectively, and outputting a load pattern vector and a fault association vector;

[0032] Performing residual connection and cross-channel fusion on the load feature, the load pattern vector, the fault feature, and the fault association vector to obtain a joint time series feature;

[0033] The joint time series features are layer normalized, and the load time series features and the fault time series features are output respectively through a dual-branch fully connected layer.

[0034] Optionally, the step of inputting the load time series characteristics and the fault time series characteristics into a fault time prediction model to predict the fault time of the current weighing sensor includes:

[0035] Constructing a time decay factor according to the working time parameter of the current weighing sensor;

[0036] The load time series features and the fault time series features are matched with each other through a time attention mechanism through a fault time prediction model to select the time series segments most relevant to the historical fault mode;

[0037] According to the time series segment and the time decay factor, a predicted failure time is output.

[0038] In addition, to achieve the above-mentioned purpose, the present application also proposes a load cell failure time prediction device, the load cell failure time prediction device comprising:

[0039] The feature extraction module is used to input the working status data of the current weighing sensor into the fault detection model to obtain candidate fault features;

[0040] a feature processing module, configured to obtain load fluctuation data through a historical load curve of the current load sensor, and preprocess the load fluctuation data and the candidate fault features to obtain load features and fault features;

[0041] A timing extraction module is used to extract timing features of the load features and the fault features based on a hybrid convolutional neural network to obtain load timing features and fault timing features;

[0042] The time prediction module is used to input the load time series characteristics and the fault time series characteristics into a fault time prediction model to predict the fault time of the current weighing sensor.

[0043] In addition, to achieve the above-mentioned purpose, the present application also proposes a weighing sensor failure time prediction device, which includes: a memory, a processor, and a computer program stored on the memory and runnable on the processor, and the computer program is configured to implement the steps of the weighing sensor failure time prediction method as described above.

[0044] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium. A computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the weighing sensor failure time prediction method as described above are implemented.

[0045] This application discloses inputting the working status data of the current weighing sensor into a fault detection model to obtain candidate fault features; obtaining load fluctuation data through the historical load curve of the current weighing sensor, and preprocessing the load fluctuation data and the candidate fault features to obtain load features and fault features; extracting time series features from the load features and the fault features based on a hybrid convolutional neural network to obtain load time series features and fault time series features; inputting the load time series features and the fault time series features into a fault time prediction model to predict the fault time of the current weighing sensor. The hybrid convolutional network is used to capture the long-term and short-term correlation between load and fault, avoiding reliance on a single feature, improving the real-time and accuracy of the prediction, and achieving end-to-end fault time prediction by integrating fault detection, time series feature extraction and prediction models, which is particularly suitable for high-fluctuation industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0048] Figure 1 This is a flow chart of the first embodiment of the load cell failure time prediction method of the present application;

[0049] Figure 2 This is a flow chart of the second embodiment of the load cell failure time prediction method of the present application;

[0050] Figure 3 This is a flow chart of the third embodiment of the load cell failure time prediction method of the present application;

[0051] Figure 4 This is a schematic diagram of the module structure of the load cell failure time prediction device according to an embodiment of the present application;

[0052] Figure 5 Schematic diagram of the device structure of the hardware operating environment involved in the weighing sensor failure time prediction method in the embodiment of the present application.

[0053] The purpose, features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0054] It should be understood that the specific embodiments described herein are merely used to explain the technical solutions of the present application and are not intended to limit the present application.

[0055] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0056] The main solution of the embodiment of the present application is: inputting the working status data of the current weighing sensor into the fault detection model to obtain candidate fault features; obtaining load fluctuation data through the historical load curve of the current weighing sensor, and preprocessing the load fluctuation data and the candidate fault features to obtain load features and fault features; extracting time series features of the load features and the fault features based on a hybrid convolutional neural network to obtain load time series features and fault time series features; inputting the load time series features and the fault time series features into the fault time prediction model to predict the fault time of the current weighing sensor.

[0057] A truck scale is a large-scale measuring instrument whose operating principle is primarily based on a combination of mechanics and electronics. When a vehicle and cargo are parked on the scale's load carrier, the force of gravity they generate is transmitted to the load cell. The elastic element in the load cell deforms under the influence of gravity, and the strain gauges attached to the elastic element also deform, causing their resistance to change. These strain gauges form a Wheatstone bridge. When the resistance of a resistor in the bridge changes, the bridge loses balance, generating an electrical signal proportional to the degree of strain. This signal is then converted to a digital signal by an A / D converter, and the weighing indicator processes this weight signal and displays it as weight data. As a core component of a truck scale, the lifecycle of the load cell is unpredictable, a major operational management challenge for many companies. Due to the lack of effective prediction methods, companies struggle to accurately assess its lifespan. This is primarily due to the complex and variable operating environment of the load cell, where its performance is affected by numerous factors such as temperature, humidity, and tilt angle. Furthermore, the load cell's manufacturing process, material selection, and usage also significantly impact its lifespan. This often leaves companies in a reactive state during daily operations, unable to develop targeted maintenance and replacement plans in advance. Once a load cell malfunctions or fails, companies often face a host of challenges, including production interruptions, quality risks, and customer complaints. Furthermore, frequent replacement of load cells increases maintenance costs, reduces equipment efficiency, and results in immeasurable economic losses. Therefore, solving the problem of load cell lifecycle prediction is crucial for ensuring stable operations, reducing operating costs, and enhancing market competitiveness.

[0058] Therefore, this application provides a method for predicting the failure time of a weighing sensor. It uses a time series model to model and predict the AD value of the weighing sensor, predict the time when the weighing sensor will fail, collect the fatigue times of the weighing sensor (maximum value and number of overloads, working time), and realize the life cycle prediction and fault diagnosis of the weighing sensor and the truck scale, so as to facilitate employees to determine the procurement and maintenance time.

[0059] It should be noted that the execution entity of this embodiment can be a computing service device with data processing, modeling analysis, and program execution functions, such as an anti-cheating monitoring platform, or an electronic device capable of implementing the above functions. This embodiment and the following embodiments are described below using a time prediction system as an example.

[0060] Based on this, the embodiment of the present application provides a method for predicting the failure time of a weighing sensor, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the load cell failure time prediction method of the present application.

[0061] In this embodiment, the load cell failure time prediction method includes:

[0062] Step S10: Input the working status data of the current weighing sensor into the fault detection model to obtain candidate fault features.

[0063] It should be noted that the current load cell operating status data is a collection of multidimensional operating parameters of the load cell collected in real time, including basic physical data (weight, temperature, humidity), environmental parameters (tilt angle, vibration frequency), and historical features (load change rate, number of overloads, baseline offset). The fault detection model is a data-driven algorithm used to identify abnormal patterns in the load cell operating status data. It can extract fault-related features (such as load fluctuations and signal mutations) from the raw data and determine the presence of potential faults through pattern matching or statistical analysis. Candidate fault features refer to fault-related characteristic parameters initially screened from the load cell operating status data by the fault detection model. Examples include the amplitude and frequency of load fluctuations, the duration of signal anomalies, and abnormal gradient changes.

[0064] It's understandable that when inputting the current load cell operating status data into the fault detection model, a data enhancement strategy combining sliding average filtering and weighted fusion can be used to effectively separate high-frequency abnormal fluctuations from low-frequency background noise, enhancing the representation of overload mutations. Compared to traditional filtering methods, this method preserves more detailed information, providing high-quality input for subsequent fault detection.

[0065] Step S20 , acquiring load fluctuation data through the historical load curve of the current weighing sensor, and preprocessing the load fluctuation data and the candidate fault features to obtain load features and fault features.

[0066] It should be noted that a historical load curve refers to a continuous data sequence of load (such as weight, pressure, etc.) changes over time, recorded by a load cell during long-term operation. This data reflects the dynamic behavior of the equipment under different operating conditions, such as peak loads, periodic fluctuations, and abnormal shocks. Load fluctuation data is a subset of the historical load curve that reflects the characteristics of load variation. This includes short-term fluctuations (such as instantaneous shocks), medium- to long-term fluctuations (such as load oscillations caused by equipment startup and shutdown), and trend changes (such as load baseline drift caused by material fatigue). Load signatures and fault features are structured representations extracted from load fluctuation data and candidate fault signatures.

[0067] As you can see, load fluctuation data can be obtained by analyzing historical load curves using a multi-scale sliding window. Time-frequency statistics are calculated within each window, and wavelet transforms are used to separate the fluctuation components in different frequency bands. Short-term windows capture transient shocks and noise interference; medium-term windows analyze the transient processes of equipment startup and shutdown or operating mode switching; and long-term windows extract load baseline drift trends.

[0068] It should be understood that when jointly preprocessing load fluctuation data and candidate fault features, an adaptive Kalman filter can be used to suppress high-frequency noise first, and timestamp alignment can be used to ensure the temporal consistency of the two types of data; then, the statistical characteristics of the load and the anomaly scores of the fault features are weightedly fused to generate a multidimensional feature vector; finally, the fused data is dynamically Z-score normalized using a sliding window to eliminate the impact of dimensional differences on the model.

[0069] Specifically, after data preprocessing, feature-layer screening is required: Physical-layer screening: Outliers (such as negative weight signals) that exceed physical probability are eliminated based on device parameters (such as rated load). Statistical-layer screening: Outliers are identified through boxplot analysis and corrected using the sliding window mean replacement method.

[0070] In one example, a truck scale load cell records the historical load curve Lt (t is the timestamp) of a truck's weight. When extracting load fluctuation data, the load standard deviation σ and peak value P for each 5-minute window are calculated:

[0071]

[0072] in, Indicates the historical load curve The measured mean of the measured values, where n is the number of load fluctuation data used in a 5-minute window.

[0073] Candidate fault characteristics include temperature drift T and signal noise During preprocessing, [σ, P, T, ] is normalized, and principal components F1 and F2 are extracted through PCA to form the final feature F = [F1, F2]. For example, when a truck is dynamically weighed, a sudden increase in σ may be associated with a loose load cell fault, while an abnormal T indicates environmental interference. After fusion, the root cause of the fault can be accurately located.

[0074] Step S30: extracting time series features from the load features and the fault features based on a hybrid convolutional neural network to obtain load time series features and fault time series features.

[0075] It should be noted that the hybrid convolutional neural network is a deep learning model that combines one-dimensional convolution (1D-CNN) and two-dimensional convolution (2D-CNN) to extract local temporal features and multidimensional spatial features from time series data, respectively. Temporal feature extraction captures dynamic patterns in data over time, such as load fluctuation trends or the periodicity of fault signals. Load temporal features are key time-correlated information extracted from load features using the hybrid CNN, such as fluctuation period and mutation frequency. Fault temporal features are temporal patterns extracted from fault features that reflect the fault's evolution, such as drift rate and noise accumulation trend.

[0076] It can be understood that the one-dimensional convolution branch in the hybrid convolutional neural network can effectively identify transient anomalies in load fluctuations (such as sudden increases or decreases), and the two-dimensional convolution branch can capture the coordinated changes of multiple fault characteristics, such as the synchronous increase in temperature and noise.

[0077] In one example, the load characteristics of a truck scale weighing sensor are the load standard deviation sequence σ=[σ1, σ2, …, σT] in a short time window, and the fault characteristics are the temperature drift T=[T1, T2, …, TT] and the noise intensity N=[N1, N2, …, NT]. 1D-CNN processes the load characteristics: 3 one-dimensional convolution kernels of size 3 are used on the σ sequence to extract the time series features. , each feature represents the fluctuation pattern at different time scales. 2D-CNN processes fault features: T and N are further stacked into a two-dimensional matrix by time , using 2×2 convolution kernel to extract spatiotemporal features .Will and After flattening, splicing , input into the fully connected layer for fault classification.

[0078] Among them, the network structure design of the hybrid CNN is as follows:

[0079] The 1D-CNN branch performs one-dimensional convolution on the load features (such as the standard deviation sequence) to capture the local dependencies in the time dimension. The formula is:

[0080]

[0081] in, is the convolution kernel weight, The first Layer features, b is the bias term, and K is the convolution kernel size.

[0082] The 2D-CNN branch combines fault characteristics (such as temperature drift and noise intensity) with the time dimension into a two-dimensional matrix and extracts spatiotemporal correlation features through two-dimensional convolution. The load timing features output by the 1D-CNN and the fault timing features output by the 2D-CNN are concatenated to form a joint timing feature vector.

[0083] Step S40: Input the load time series characteristics and the fault time series characteristics into a fault time prediction model to predict the fault time of the current weighing sensor.

[0084] It should be noted that the failure time prediction model is a machine learning model based on time series data. It analyzes the time series evolution of load and fault characteristics to predict the remaining useful life (RUL) of the load cell or the specific time when a failure will occur. The failure time prediction is the time point at which the load cell is likely to fail in the future or the remaining service life.

[0085] In one example, the fault time prediction model uses an encoder-decoder architecture, such as the LSTM-Transformer hybrid model. First, feature encoding is performed to convert the load time series features into and fault timing characteristics Input into LSTM encoder separately to capture long-term dependencies:

[0086]

[0087] Among them, T is the number of time steps, d1, d2 are feature dimensions, is the hidden state at step t.

[0088] Next, perform time prediction decoding to convert the final hidden state Input the fully connected layer and output the failure time through regression :

[0089]

[0090] Where W is the weight matrix and b is the bias term. If probabilistic prediction is required, a survival analysis model can be used instead.

[0091] In this embodiment, the working status data of the current weighing sensor is input into the fault detection model to obtain candidate fault features; load fluctuation data is obtained through the historical load curve of the current weighing sensor, and the load fluctuation data and the candidate fault features are preprocessed to obtain load features and fault features; time series features are extracted from the load features and the fault features based on a hybrid convolutional neural network to obtain load time series features and fault time series features; the load time series features and the fault time series features are input into the fault time prediction model to predict the fault time of the current weighing sensor. The hybrid convolutional network is used to capture the long-term and short-term associations between load and fault, avoiding reliance on a single feature, improving the real-time and accuracy of the prediction, and by integrating fault detection, time series feature extraction and prediction models, end-to-end fault time prediction is achieved, which is particularly suitable for high-volatility industrial scenarios.

[0092] Reference Figure 2 , Figure 2 This is a flow chart of the second embodiment of the method for predicting the failure time of a weighing sensor of the present application. Based on the above-mentioned first embodiment, the second embodiment of the method for predicting the failure time of a weighing sensor of the present application is proposed.

[0093] In the second embodiment, step S10 includes:

[0094] Step S101 : Processing the working state data of the current weighing sensor through a sliding average filter template to obtain first state data.

[0095] It should be noted that the sliding average filter template is a mathematical model used for signal smoothing. It calculates a weighted average of the load cell operating status data within a fixed time window to suppress high-frequency noise and extract low-frequency trend components. The first state data is the output of the raw load cell operating status data after sliding average filtering, representing the low-frequency trend component of the data.

[0096] Specifically, a sliding window of fixed length (e.g., the past N sampling points) is set. The window moves forward as new data is added. Equal weighting (simple sliding average) or adaptive weighting (e.g., Gaussian weighting or exponential decay) is used to assign different weight coefficients to the data within the window, highlighting the importance of recent data. The calculation method is to superimpose the weighted data within the window and then normalize them, outputting the filtering result at the current moment.

[0097] It should be understood that by separating high-frequency abnormal fluctuations (such as transient overload) from low-frequency background noise (such as temperature drift), key mutation characteristics can be preserved. Compared with traditional filtering methods (such as low-pass filtering), the dynamic weighting strategy reduces detail loss and provides high-quality input for subsequent fault detection.

[0098] Step S102: removing the first status data from the working status data to obtain second status data.

[0099] It should be noted that the second-state data is the high-frequency component separated from the original operating-state data, i.e., the residual between the original data and the first-state data (the low-frequency trend component). By removing instantaneous fluctuations or abnormal signals (such as transient shocks and sudden noise) from the long-term stable background, the short-term abnormalities or fault characteristics of the load cell are captured.

[0100] It can be understood that compared with the first state data, the second state data has a greater weight in fault time prediction by eliminating low-frequency interference (such as temperature drift and slow wear) and amplifying high-frequency mutation signals.

[0101] Step S103: The second state data and the working state data are weightedly fused to obtain third state data.

[0102] It should be noted that the third state data is the result of weighted fusion of the original working state data and the second state data. It has both the integrity of the original data and the high-frequency abnormality sensitivity of the second state data, and is used to balance noise suppression and detail retention.

[0103] It is understandable that the coefficients of the fusion weight can be adaptively adjusted based on prior knowledge or signal characteristics. For example, if the dominance of the original data needs to be emphasized, the weight of the working status data can be increased. By adjusting the weight, the high-frequency mutations related to the fault can be retained or enhanced while suppressing background noise (such as temperature drift).

[0104] Step S104: input the third state data into a preset fault detection model to obtain candidate fault features.

[0105] Furthermore, in order to reduce misjudgments caused by environmental interference through adaptive noise suppression and local anomaly discrimination, so as to maintain a high detection recall rate under complex working conditions, the step S104 may include:

[0106] The candidate fault interval of the third state data is determined by an improved density peak search method; local abnormality judgment is performed on the candidate fault interval to generate a fault probability distribution map; noise is suppressed on the fault probability distribution map by an adaptive filtering method to obtain a suppressed fault interval, and a target fault interval corresponding to the second state data is determined based on the suppressed fault interval; dynamic threshold segmentation is performed on the target fault interval to obtain candidate fault features.

[0107] It should be noted that the improved density peak search method is an optimization algorithm based on density clustering. It automatically identifies high-density areas (candidate fault intervals) by calculating the local density and relative distance of data points. Improvements include adaptive neighborhood radius and the introduction of timing constraints to adapt to the spatiotemporal characteristics of weighing sensor data. Density peak search is used to preliminarily identify candidate fault intervals, identifying time periods or data intervals that may contain faults. A fault probability distribution diagram is a two-dimensional distribution diagram with time or space as the horizontal axis and probability value as the vertical axis, visually displaying the probability of faults at different locations.

[0108] Additionally, it's important to note that adaptive filtering dynamically adjusts filtering parameters based on the noise characteristics of the probability distribution graph, suppressing random noise while retaining the fault signal. Fault interval after suppression: The continuous region in the denoised probability distribution graph where the probability value exceeds a preset threshold is considered a more reliable fault candidate.

[0109] It's understandable that adaptive filtering methods require dynamic adjustment of filtering parameters based on the noise characteristics of the probability distribution graph, suppressing random noise while retaining the fault signal. When performing dynamic threshold segmentation, the segmentation threshold can be adjusted in real time based on the data distribution to extract candidate fault features from the target fault interval.

[0110] In one example, the third state data , calculate each time point The local density , the formula is:

[0111]

[0112] in for and The Euclidean distance of The neighborhood radius is adaptive and can be adjusted dynamically according to the data sparsity. Calculate the minimum distance δi between each point and the point with higher density, and select The point with the largest product with δi is considered the density peak. Peak points must appear at least multiple times consecutively on the time axis to avoid misidentifying isolated noise as a fault. Adjacent density peak points are merged into candidate fault intervals.

[0113] When generating the fault probability distribution diagram, for each data point in the candidate interval , define the time series neighborhood window, and then use the local outlier factor (LOF) algorithm to calculate the anomaly score :

[0114]

[0115] in is the local reachable density, N(i) is the neighborhood point set, and finally Normalize to probability value and generate fault probability distribution map. Select probability threshold and extract continuous intervals with fault probability greater than probability threshold as suppressed fault intervals.

[0116] Furthermore, in order to calculate the load density peak based on the combined weight of the maximum overload value and the number of overloads, the limitation of the traditional method of relying on only a single indicator is overcome. By sorting the probability within the time window, high-risk periods can be quickly identified, improving the efficiency of screening candidate fault intervals. The step of determining the candidate fault interval of the third state data by improving the density peak search method can include:

[0117] Extract the maximum overload value and the number of overloads at each time point in the third state data; calculate the load density at the time point based on the number of overloads at adjacent time points; determine the load density peak point through a joint weight based on the maximum overload value and the load density; construct a time window to be detected with the load density peak point as the center, calculate the overload probability value within the time window to be detected, and sort the overload probability values ​​in descending order to obtain the candidate fault interval.

[0118] It should be noted that the maximum overload value refers to the maximum instantaneous value at which the load cell's measured value exceeds the rated load at a specific point in time. The number of overloads is the cumulative number of times the load cell's measured value exceeds the rated load threshold within a specified time window. Load density indicates the frequency of overload events around a specific point in time and is calculated by weighted accumulation of overloads at adjacent time points.

[0119] The load density peak is the time point with the highest overall score after calculating the combined weights of load density and maximum overload, representing the location with the highest failure risk. The detection time window is a fixed time interval centered around the load density peak. The overload probability is the ratio of the number of overload events occurring within the time window to the total number of sampling points, reflecting the probability of failure within that interval.

[0120] In one example, extracting the maximum overload value and the number of overloads at each time point in the third state data includes: Determine whether it is overloaded point by point ( , is the rated load). Record the overload amount at the overload point. , and retain the maximum value at that time point, and count the cumulative number of overloads within the window to which each time point belongs .

[0121] Load density requires time points , aggregate the overload times of its adjacent k points (such as 5 points before and after), and calculate the load density :

[0122]

[0123] in, is the distance attenuation weight, such as =1 / (d+1), d is the time interval.

[0124] Then Normalization is performed and the comprehensive score is calculated according to the weight coefficients α (overload maximum value weight) and β (load density weight):

[0125]

[0126] in, The point where the score exceeds the threshold and is the local maximum is selected as the load density peak point. .

[0127] Finally, the time window is constructed and the overload probability is calculated, with the peak point As the center, expand time T to both sides to build a window , and count the percentage of overload times within the window , press all windows Arrange in descending order and select the top M (e.g., top 10%) as candidate fault intervals.

[0128] Furthermore, in order to introduce directional consistency factors and multi-directional fluctuation conditions to effectively distinguish random fluctuations from real fault signals, the dynamic weighting of gradient deviation and load deviation is combined to enhance sensitivity to hidden faults (such as gradual aging) and reduce the risk of missed detection. The step of performing local abnormality identification on the candidate fault interval and generating a fault probability distribution map may include:

[0129] Obtain the first load mean and the first gradient mean of the current time window in the candidate fault interval, as well as the second load mean and the second gradient mean of the adjacent window; determine the load deviation based on the first load mean and the second load mean, and determine the gradient mean difference based on the first gradient mean and the second gradient mean; determine the directional consistency factor based on the positive and negative signs of the gradient mean difference; if the directional consistency factor meets the multi-directional fluctuation condition, mark the candidate fault interval as a normal interval; if the directional consistency factor does not meet the multi-directional fluctuation condition, generate a fault probability distribution diagram based on the load deviation, the gradient mean difference and the directional consistency factor.

[0130] It should be noted that the first load mean is the arithmetic mean of the load data of the weighing sensor in the current time window within the candidate fault interval; the first gradient mean is the average value of the instantaneous change rate of the load data in the current time window; the second load mean is the load mean of another time window adjacent to the current window; the second gradient mean is the average value of the load change rate in the adjacent windows.

[0131] Additionally, it's important to note that load deviation is the difference between the load mean of the current window and the adjacent windows, and is used to quantify the load level offset. Gradient mean difference is the difference between the gradient mean of the current window and the adjacent windows, reflecting the difference in load change trends.

[0132] The directional consistency factor is a factor based on the sign (positive / negative) of the gradient mean difference, used to determine whether the load fluctuation direction is consistent. The multi-directional fluctuation condition is a condition where the gradient direction (positive / negative) of adjacent windows alternates, used to distinguish normal fluctuations from faults.

[0133] In one example, the current window is a subwindow within the candidate fault interval, and the adjacent window is a window of the same length that is adjacent to the current window. The first and second load means are calculated as follows:

[0134]

[0135] Where S(t) is the load data of the weighing sensor and N is the number of sampling points in the window. The first and second gradient means are calculated as follows:

[0136]

[0137] in Load data for adjacent windows.

[0138] The load deviation and gradient mean difference are:

[0139] ,

[0140] like , the directional consistency factor sign = +1; otherwise sign = −1. High load deviations with consistent gradient directions (e.g., all positive) may indicate sustained overload; inconsistent directions indicate normal fluctuations.

[0141] When the sign of the directional consistency factor is stable, the failure probability calculation is performed:

[0142]

[0143] in, and is the historical maximum value, and finally the time window is used as the horizontal axis. As the vertical axis, draw a probability heat map.

[0144] In this embodiment, the current load cell operating status data is processed using a sliding average filter template to obtain first status data. The first status data is then removed from the operating status data to obtain second status data. The second status data is then weightedly fused with the operating status data to obtain third status data. This third status data is then input into a pre-set fault detection model to obtain candidate fault signatures. This data enhancement strategy, combining sliding average filtering with weighted fusion, effectively separates high-frequency abnormal fluctuations from low-frequency background noise, enhancing the representation of overload mutation signatures. Compared to traditional filtering methods, this method preserves more detailed information, providing high-quality input for subsequent fault detection.

[0145] Reference Figure 3 , Figure 3 This is a flow chart of the third embodiment of the method for predicting the failure time of a weighing sensor of the present application. Based on the above second embodiment, the third embodiment of the method for predicting the failure time of a weighing sensor of the present application is proposed.

[0146] In the third embodiment, step S30 includes:

[0147] Step S301: input the load characteristics and the fault characteristics into the convolutional layers of a hybrid convolutional neural network respectively, and output a load pattern vector and a fault association vector.

[0148] It should be noted that the load pattern vector is a high-dimensional vector representation of the load feature after being mapped by the convolution layer; the fault association vector is a high-dimensional vector representation of the fault feature after being mapped by the convolution layer.

[0149] It can be understood that by processing load and fault features through independent convolution branches, feature confusion can be avoided, and the wide / narrow convolution kernel combination can simultaneously capture the long-term attenuation trend of the load (such as metal fatigue) and the short-term mutation of the fault.

[0150] It should be understood that after the load signature is input into the convolutional layer, the convolution kernel slides over the load signature data, performing convolution calculations on each local region. Specifically, the convolution kernel is multiplied by the corresponding elements of the local region and the sum is calculated to obtain a new value, which becomes an element of the convolution output. By continuously sliding the convolution kernel, the convolution output of the entire load signature is obtained. The convolution output of the fault signature is also subjected to activation functions and pooling operations to obtain a fault association vector. This vector can reflect the correlation between faults, such as whether certain faults often occur together and the order in which they occur.

[0151] Step S302 : performing residual connection and cross-channel fusion on the load feature, the load pattern vector, the fault feature, and the fault association vector to obtain a joint time series feature.

[0152] It should be understood that residual connections and cross-channel fusion can deeply integrate the original features with the abstract features extracted by the convolutional layer to generate more discriminative joint temporal features. In residual connections, the features and the corresponding vector inputs are first aligned, and then residual superposition is performed. In cross-channel fusion, the residual outputs of the load and fault branches are spliced ​​into multi-channel features, weightedly fused using channel attention weights, and then cross-channel information integration is performed using 1×1 convolution kernels.

[0153] Step S303 , performing layer normalization on the joint time series features, and outputting the load time series features and the fault time series features respectively through a dual-branch fully connected layer.

[0154] As you can understand, the dual-branch fully connected layer processes load and fault features separately through two independent fully connected sub-networks, extracting their respective timing patterns. This avoids information interference between load and fault features in the shared network and allows for the optimization of dedicated features for different tasks (such as load prediction and fault diagnosis).

[0155] In the third embodiment, step S40 includes:

[0156] Step S401: constructing a time decay factor according to the working time parameter of the current weighing sensor.

[0157] It should be noted that the time decay factor is a weighting coefficient based on the load cell's cumulative operating time or recent usage intensity. It is used to quantify the cumulative effect of equipment aging or fatigue on failure risk. It can be designed as a monotonically decreasing function, reflecting the decreasing predictive contribution of historical failure modes over time.

[0158] In one example, the input working time parameters include the accumulated working time and recent use intensity (The average load rate in the most recent time window). The decay function is designed to be exponential decay:

[0159]

[0160] in, is the attenuation rate parameter, obtained by fitting historical fault data. Then, the attenuation factor is corrected in combination with the recent usage intensity:

[0161]

[0162] in, is the strength sensitivity coefficient, which is determined based on the material of the weighing sensor. >1 indicates overload and accelerated decay.

[0163] Step S402: Match the load time series features and the fault time series features with a time attention mechanism through a fault time prediction model to select the time series segment most relevant to the historical fault mode.

[0164] It should be noted that the temporal attention mechanism matching is a dynamic weight allocation method that screens out the most relevant temporal segments for the current prediction by calculating the degree of temporal alignment between the current temporal features and the historical fault modes.

[0165] In one example, before matching, the load timing characteristics need to be and fault timing characteristics Perform feature alignment to obtain the splicing vector , the fault mode characteristics stored in the historical fault library are (N is the number of historical failures, D is the feature dimension). For each historical failure mode , calculate the time alignment score with the current feature as follows:

[0166]

[0167] Then, the time segment corresponding to the historical fault mode with the highest weight is selected.

[0168] Step S403: outputting a predicted failure time according to the time sequence segment and the time attenuation factor.

[0169] It can be understood that statistical features (such as mean, variance, and slope) can be extracted from the filtered time series segments to generate feature vectors, and the time decay factor is used as the weight coefficient to correct the prediction weight of the historical failure mode. At the same time, the fused features are input into the fully connected regression layer to output the failure time.

[0170] In one example, the filtered time segment Extract statistical features (such as mean, variance, slope) and generate feature vectors , the time decay factor As a weight coefficient, the prediction weight of the historical failure mode is modified:

[0171]

[0172] Input fusion features To the fully connected regression layer, output failure time :

[0173]

[0174] in, is the regression weight matrix, is the regression bias term. Of course, the failure risk curve can also be combined with the survival analysis model to quantify the failure time.

[0175] In this embodiment, a hybrid convolutional network integrates residual connections and cross-channel interactions to simultaneously extract local anomalies and long-range dependency features. Layer normalization and dual-branch outputs address the scale differences between load and fault features, improving the model's ability to model complex time series patterns. By matching historical failure patterns and dynamically adjusting prediction weights, predictions are more closely aligned with actual service life, making them particularly suitable for lifespan prediction of high-load equipment.

[0176] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the weighing sensor failure time prediction method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0177] This application also provides a load cell failure time prediction device, please refer to Figure 4 , the load cell failure time prediction device comprises:

[0178] The feature extraction module 10 is used to input the working state data of the current weighing sensor into the fault detection model to obtain candidate fault features;

[0179] a feature processing module 20 for acquiring load fluctuation data through a historical load curve of the current load cell, and preprocessing the load fluctuation data and the candidate fault features to obtain load features and fault features;

[0180] A time series extraction module 30 is configured to extract time series features from the load features and the fault features based on a hybrid convolutional neural network to obtain load time series features and fault time series features;

[0181] The time prediction module 40 is configured to input the load time series characteristics and the fault time series characteristics into a fault time prediction model to predict the fault time of the current weighing sensor.

[0182] The load cell failure time prediction device provided in this application utilizes the load cell failure time prediction method described in the aforementioned embodiment, addressing the technical issue in the prior art of a lack of effective means for predicting the lifecycle of load cells, which prevents companies from formulating targeted maintenance and replacement plans in advance. Compared to the prior art, the load cell failure time prediction device provided in this application has the same beneficial effects as the load cell failure time prediction method described in the aforementioned embodiment, and the other technical features of the load cell failure time prediction device are the same as those disclosed in the aforementioned embodiment, which are not further detailed here.

[0183] The present application provides a weighing sensor failure time prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the weighing sensor failure time prediction method in the above-mentioned embodiment one.

[0184] Reference below Figure 5 , which shows a schematic diagram of the structure of a load cell failure time prediction device suitable for implementing an embodiment of the present application. The load cell failure time prediction device in the embodiment of the present application can include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (such as in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The load cell failure time prediction device shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0185] like Figure 5As shown, the load cell failure time prediction device may include a processing device 1001 (e.g., a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in a read-only memory 1002 or programs loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the load cell failure time prediction device. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: an input device 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image load cell, microphone, accelerometer, gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; a storage device 1003 including, for example, a magnetic tape or hard disk; and a communication device 1009. The communication device 1009 can allow the load cell failure time prediction device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a load cell failure time prediction device with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented or have alternatively.

[0186] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are performed.

[0187] The load cell failure time prediction device provided in this application, which utilizes the load cell failure time prediction method described in the aforementioned embodiment, can address the technical issue in the prior art of lacking effective means for predicting the lifecycle of load cells, preventing companies from formulating targeted maintenance and replacement plans in advance. Compared to the prior art, the load cell failure time prediction device provided in this application has the same beneficial effects as the load cell failure time prediction method described in the aforementioned embodiment. The other technical features of the load cell failure time prediction device are the same as those disclosed in the aforementioned embodiment and are not further elaborated here.

[0188] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0189] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0190] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, a computer program) stored thereon, wherein the computer-readable program instructions are used to execute the weighing sensor failure time prediction method in the above-mentioned embodiment.

[0191] The computer-readable storage medium provided herein may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including, but not limited to, wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0192] The computer-readable storage medium may be included in the load cell failure time prediction device; or may exist independently without being assembled into the load cell failure time prediction device.

[0193] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the weighing sensor failure time prediction device, the weighing sensor failure time prediction device executes the weighing sensor failure time prediction method described above.

[0194] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0195] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of code, and the module, program segment or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0196] The modules described in the embodiments of the present application may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.

[0197] The computer-readable storage medium provided in this application stores computer-readable program instructions (i.e., a computer program) for executing the aforementioned load cell failure time prediction method. This computer-readable storage medium addresses the technical issue of the prior art lacking effective methods for predicting the lifecycle of load cells, which prevents companies from formulating targeted maintenance and replacement plans in advance. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are similar to those of the load cell failure time prediction method provided in the aforementioned embodiments and are not further elaborated here.

[0198] The above description is only part of the embodiments of the present application and does not limit the patent scope of the present application. All equivalent structural transformations made by using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for predicting the failure time of a weighing sensor, characterized in that: The load cell failure time prediction method comprises: Input the working status data of the current weighing sensor into the fault detection model to obtain candidate fault features; Acquire load fluctuation data through a historical load curve of the current load cell, and pre-process the load fluctuation data and the candidate fault features to obtain a load feature and a fault feature; Performing time series feature extraction on the load feature and the fault feature based on a hybrid convolutional neural network to obtain a load time series feature and a fault time series feature; Inputting the load time series characteristics and the fault time series characteristics into a fault time prediction model to predict the fault time of the current weighing sensor; The step of inputting the working state data of the current weighing sensor into the fault detection model to obtain candidate fault features includes: Processing the working state data of the current weighing sensor through a sliding average filter template to obtain first state data; removing the first state data from the working state data to obtain second state data; The second state data and the working state data are weightedly fused to obtain third state data; Inputting the third state data into a preset fault detection model to obtain candidate fault features; The step of inputting the third state data into a preset fault detection model to obtain candidate fault features includes: determining a candidate fault interval of the third state data by an improved density peak search method; Performing local abnormality discrimination on the candidate fault interval to generate a fault probability distribution map; performing noise suppression on the fault probability distribution graph by an adaptive filtering method to obtain a suppressed fault interval, and determining a target fault interval corresponding to the second state data based on the suppressed fault interval; Performing dynamic threshold segmentation on the target fault interval to obtain candidate fault features; The step of determining the candidate fault interval of the third state data by using the improved density peak search method includes: Extracting the maximum overload value and the number of overload times at each time point in the third state data; Calculating the load density at the time point based on the number of overloads at adjacent time points; Determine a load density peak point based on the maximum overload value and the load density by combining weights; A time window to be detected is constructed with the load density peak point as the center, an overload probability value within the time window to be detected is calculated, and the overload probability values ​​are sorted in descending order to obtain the candidate fault interval.

2. The load cell failure time prediction method according to claim 1, wherein: The step of performing local abnormality identification on the candidate fault interval and generating a fault probability distribution map includes: Obtaining a first load mean and a first gradient mean of a current time window in the candidate fault interval, and a second load mean and a second gradient mean of an adjacent window; determining a load deviation according to the first load mean and the second load mean, and determining a gradient mean difference according to the first gradient mean and the second gradient mean; Determining a directional consistency factor based on the positive and negative signs of the gradient mean difference; If the directional consistency factor satisfies the multi-directional fluctuation condition, marking the candidate fault interval as a normal interval; If the directional consistency factor does not meet the multi-directional fluctuation condition, a fault probability distribution diagram is generated according to the load deviation, the gradient mean difference and the directional consistency factor.

3. The method for predicting the failure time of a weighing sensor according to any one of claims 1 to 2, characterized in that: The step of extracting time series features from the load features and the fault features based on a hybrid convolutional neural network to obtain load time series features and fault time series features includes: Inputting the load feature and the fault feature into the convolution layer of the hybrid convolutional neural network respectively, and outputting a load pattern vector and a fault association vector; Performing residual connection and cross-channel fusion on the load feature, the load pattern vector, the fault feature, and the fault association vector to obtain a joint time series feature; The joint time series features are layer normalized, and the load time series features and the fault time series features are output respectively through a dual-branch fully connected layer.

4. The method for predicting failure time of a weighing sensor according to any one of claims 1 to 2, characterized in that: The step of inputting the load time series characteristics and the fault time series characteristics into a fault time prediction model to predict the fault time of the current weighing sensor includes: Constructing a time decay factor according to the working time parameter of the current weighing sensor; The load time series features and the fault time series features are matched with each other through a time attention mechanism through a fault time prediction model to select the time series segments most relevant to the historical fault mode; According to the time series segment and the time decay factor, a predicted failure time is output.

5. A device for predicting the failure time of a weighing sensor, characterized in that: The device comprises: The feature extraction module is used to input the working status data of the current weighing sensor into the fault detection model to obtain candidate fault features; a feature processing module, configured to obtain load fluctuation data through a historical load curve of the current load sensor, and preprocess the load fluctuation data and the candidate fault features to obtain load features and fault features; A timing extraction module is used to extract timing features of the load features and the fault features based on a hybrid convolutional neural network to obtain load timing features and fault timing features; A time prediction module, configured to input the load time series characteristics and the fault time series characteristics into a fault time prediction model to predict the fault time of the current weighing sensor; The feature extraction module is further configured to process the working state data of the current weighing sensor using a sliding average filter template to obtain first state data; remove the first state data from the working state data to obtain second state data; perform weighted fusion of the second state data and the working state data to obtain third state data; and input the third state data into a preset fault detection model to obtain candidate fault features. The feature extraction module is further configured to determine candidate fault intervals of the third state data using an improved density peak search method; perform local abnormality discrimination on the candidate fault intervals to generate a fault probability distribution map; perform noise suppression on the fault probability distribution map using an adaptive filtering method to obtain a suppressed fault interval, and determine a target fault interval corresponding to the second state data based on the suppressed fault interval; perform dynamic threshold segmentation on the target fault interval to obtain candidate fault features; The feature extraction module is also used to extract the maximum overload value and the number of overloads at each time point in the third state data; calculate the load density at the time point based on the number of overloads at adjacent time points; determine the load density peak point through a joint weight based on the maximum overload value and the load density; construct a time window to be detected with the load density peak point as the center, calculate the overload probability value within the time window to be detected, and sort the overload probability values ​​in descending order to obtain the candidate fault interval.

6. A load cell failure time prediction device, characterized in that: The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the load cell failure time prediction method according to any one of claims 1 to 4.

7. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the weighing sensor failure time prediction method according to any one of claims 1 to 4 are implemented.

Images