An intelligent maintenance method for small-parts sorting machines in logistics transfer yards based on digital twins

Abstract

The present invention discloses a digital twin-based intelligent maintenance method for small-piece sorting machines in logistics transfer yards, comprising the following steps: S1. Deploying multiple sensors at key locations on the cross-belt sorting machine to collect data in real time, transmitting the data to a preprocessing module for preprocessing to obtain preprocessed data vectors, and constructing a data matrix within a time window; S2. Constructing a digital twin model of the cross-belt sorting machine; S3. Calculating the cumulative health index and remaining life of the cross-belt sorting machine based on historical data, and then constructing a training set based on fault status, cumulative health index, and remaining life; S4. Constructing a CNN model to predict fault status, cumulative health index, and remaining life, optimizing the CNN model, and calculating a warning index to specify a maintenance strategy. This invention, while balancing real-time data collection, digital twins, and deep learning intelligent maintenance, achieves dynamic health monitoring, fault prediction, and maintenance warnings for small-piece sorting machines in logistics transfer yards.

Description

Technical Field

[0001] The present invention relates to the field of logistics sorting, and in particular to an intelligent maintenance method for small-parts sorting machines in logistics transfer yards based on digital twins. Background Art

[0002] In the current logistics industry, small-item sorting machines are crucial equipment in transit hubs. Their efficient and stable operation is directly linked to logistics efficiency and equipment maintenance costs. Traditional maintenance models for small-item sorting machines rely primarily on periodic inspections or post-event maintenance, which suffers from the following deficiencies: 1. Insufficient real-time monitoring: Key equipment components (such as the transmission system and control unit) lack real-time monitoring, which can easily lead to the accumulation of hidden dangers. 2. Data silos: Data collected by various sensors, PLC controllers, and cameras lacks effective integration, making it difficult to form health indicators for the entire equipment lifecycle. 3. Incomplete predictive models: Traditional empirical maintenance methods struggle to predict faults based on equipment operating status, historical fault data, and environmental factors, resulting in inaccurate equipment maintenance warnings. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide an intelligent maintenance method for small-part sorting machines in logistics transfer yards based on digital twins. On the basis of taking into account real-time data acquisition, digital twins and deep learning intelligent maintenance, dynamic health monitoring, fault prediction and maintenance warning of small-part sorting machines in logistics transfer yards are realized.

[0004] The purpose of the present invention is to achieve the following technical solution: a digital twin-based intelligent maintenance method for small item sorting machines in logistics transfer yards, comprising the following steps:

[0005] S1. Deploy multiple sensors at key locations on the cross-belt sorter to collect data in real time, transmit it to the preprocessing module for preprocessing, obtain preprocessed data vectors, and construct a data matrix within the time window;

[0006] S2. Build a digital twin model of the cross-belt sorter;

[0007] S3. Calculate the cumulative health index and remaining life of the cross-belt sorter based on historical data, and then construct a training set based on fault status, cumulative health index, and remaining life;

[0008] S4. Construct a CNN model to predict fault status, cumulative health index, and remaining life, optimize the design of the CNN model, and calculate the early warning index to specify the maintenance strategy.

[0009] The beneficial effects of the present invention are as follows: the present invention realizes the real-time collection and preprocessing of the equipment operation status by deploying multiple sensors, PLC controllers, cameras and key monitoring equipment such as vibration, temperature, operating time, current, and noise: Specifically, the present invention establishes a digital twin model driven by multi-source real-time data, realizes two-way data synchronization between physical equipment and virtual models, dynamically restores the mechanical structure, control logic and environmental impact of the equipment, and quantitatively describes the health status of the equipment through a full life cycle health assessment model, revealing the inherent coupling relationship between the operating status of key parts of the equipment, health index and potential fault hazards;

[0010] To further achieve intelligent maintenance, the present invention introduces a fault diagnosis and health status assessment method based on a convolutional neural network (CNN). This method automatically learns features from preprocessed multidimensional sensor data by constructing a deep network structure that incorporates spatiotemporal feature extraction and an attention mechanism, establishing a multi-task model for fault status, cumulative health index, and remaining life prediction. This system can thus accurately diagnose equipment status, accurately predict potential failure risks, and quantitatively assess remaining life. In summary, the present invention, while integrating real-time data acquisition, digital twins, and deep learning intelligent maintenance, achieves dynamic health monitoring, fault prediction, and maintenance warnings for small-piece sorting machines in logistics transfer stations. This overcomes the technical bottleneck of traditional periodic maintenance models, which suffer from insufficient perception of equipment health status and delayed fault prediction, and provides new theoretical basis and technical support for the field of intelligent logistics equipment maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0012] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0013] With the development of the Internet of Things, digital twins, and deep learning technologies, by establishing a two-way data synchronization mechanism between physical devices and virtual models, not only can the status of the equipment be restored in real time, but also multi-source data can be integrated and analyzed with the help of deep learning algorithms to achieve fault prediction, remaining life estimation, and intelligent maintenance scheduling. Specifically, the entire method of the present invention is divided into three levels: physical layer: including actual sorting equipment, sensor networks, PLC controllers, etc. for data collection; virtual modeling layer: through functions Restore the device status and form a virtual device model; the intelligent analysis layer, namely CNN: embedded in the digital twin system, is used to automatically learn device status characteristics from multi-dimensional data to achieve prediction and decision support;

[0014] like Figure 1As shown in FIG, a digital twin-based intelligent maintenance method for a small-parts sorting machine in a logistics transfer station includes the following steps:

[0015] S1. Deploy multiple sensors at key locations on the cross-belt sorter to collect data in real time, transmit it to the preprocessing module for preprocessing, obtain preprocessed data vectors, and construct a data matrix within the time window;

[0016] Deploy various sensors at key locations of the cross-belt sorter, such as the conveyor belt, sorting grid, and scanner, to collect real-time information such as vibration, temperature, operating time, current, and noise. Set the value sampled by each sensor at time t to , the collected sensor data is formed into a preprocessed data vector:

[0017]

[0018] Where m is the number of sensors. We use PLC controllers and cameras to transmit data to a data preprocessing module via industrial Ethernet or other fieldbuses. This module denoises and normalizes the raw data to form a data matrix suitable for subsequent modeling.

[0019] Based on the preprocessed data vector, construct a data matrix within the time window:

[0020] Where T is the length of the time window.

[0021] S2. Build a digital twin model of the cross-belt sorter;

[0022] According to the CAD model, kinematic and dynamic parameters of the actual cross-belt sorter, an accurate three-dimensional model of the cross-belt sorter is established, and the pre-processed data vector is converted into Import the digital twin model to synchronize the sorting machine digital twin model with actual data.

[0023] In some embodiments, feedback control can also be performed based on the digital twin model:

[0024] Defining a function F The expression combines the real-time data collected by sensors with physical, control and environmental models to achieve dynamic restoration of equipment status.

[0025]

[0026] in, represents the state of the virtual model at time t (including mechanical structure, control logic and environmental state), It is a collection of physical parameters, control parameters and environmental impact parameters.

[0027] To ensure the consistency between the digital twin model and the physical device status, the system establishes a two-way data synchronization mechanism:

[0028] On the one hand, the preprocessed data vector Import the digital twin model to synchronize the sorting machine digital twin model with the actual data; on the other hand, observe whether the actual running status of the digital twin model is consistent with the expected state. If it is inconsistent, it needs to be adjusted. The control parameters are used to realize feedback control;

[0029] After the digital twin model is built, it can be used to display the operation status of the sorting machine for staff to monitor in real time.

[0030] S3. Calculate the cumulative health index and remaining life of the cross-belt sorter based on historical data, and then construct a training set based on fault status, cumulative health index, and remaining life;

[0031] S301. Obtain data in multiple time windows according to step S1. The data in each time window is recorded as:

[0032]

[0033] S302. For data within any time window, calculate the cumulative health index and estimate the remaining lifespan, including:

[0034] Set the device instantaneous health index function , the cumulative health index is obtained by integrating the time :

[0035]

[0036] in, For the mapping function:

[0037]

[0038] in, Represents the mapping function input variable, It is an adjustment parameter based on human experience, and the response sensitivity between the health value and the actual equipment degradation degree is controlled by historical sorting performance;

[0039] Used to reflect the nonlinear relationship between the cumulative health index and the overall status of the equipment Expressed as:

[0040]

[0041] in For the i The weight of each sensor feature; ;

[0042] According to the cumulative health index and the preset health threshold , estimated remaining life :

[0043]

[0044] in, is the equipment health degradation rate, obtained by fitting historical data and regarded as a known value;

[0045] S303. Take any time window data as sample feature and add cumulative health index , Remaining life and as a label; is the fault status label, : , indicating that the cross-belt sorter has experienced a fault within the time window; , indicating that the cross-belt sorter has no faults within the time window;

[0046] The training sample of time window data is recorded as ;

[0047] S304. For each time window data, repeat steps S302 to S303 to obtain the training samples corresponding to each time window data, add them to the same set to form a training set.

[0048] In some embodiments, the samples may be divided into a training set and a test set. After the CNN model is trained using the training set, the accuracy of the model may be tested using the test set.

[0049] Traditional health assessment function methods have problems such as modeling difficulties and complex parameter adjustment when facing complex working conditions and multi-source heterogeneous data. To address this, this paper introduces a data-driven method based on CNN to directly learn the equipment's fault status, cumulative health index, and remaining life prediction model from sensor data. Specifically, CNN can be regarded as a function The approximate fitter outputs a cumulative health index Alternative to traditional Real-time monitoring; at the same time, CNN output It can be used as the estimated value of remaining life, avoiding the complex integration and mapping process. The above two model paths can be flexibly selected according to the actual application needs: in the model training stage, the traditional health status function is used and It can serve as a pseudo-label source for CNN; during the operation phase, the CNN output results can be used as the leading indicator, while the traditional parsing path can be used as an auxiliary analysis tool or model interpretation method, thereby realizing a "data-driven + knowledge modeling" integrated health assessment architecture.

[0050] S4. Construct a CNN model to predict fault status, cumulative health index, and remaining life, optimize the design of the CNN model, and calculate the early warning index to specify the maintenance strategy.

[0051] S401. Input Data Tensor Construction: After the data preprocessing module, the real-time sensor data is normalized, denoised, and windowed to form the following data matrix:

[0052]

[0053] Reconstruct the data into a three-dimensional tensor format, set in the form of "height-width-channel", where the height Corresponding to the number of deployed sensors, width Corresponding time step, number of channels According to the specific fusion method, each sensor data constitutes a channel separately. Represents the feature channel corresponding to each sensor; the final input tensor is recorded as ;

[0054] S402. Randomly crop the input tensor along the time axis, inject noise to simulate sensor errors, and randomly scale the input tensor to expand the dataset, allowing the model to better adapt to data variations in actual work.

[0055] S403. By stacking multiple 2D convolutional layers, the model gradually captures features from local to global. In the stacked 2D convolutional layers, the output feature map of the previous 2D convolutional layer serves as the input feature map of the next 2D convolutional layer. The input feature map of the first 2D convolutional layer serves as the input tensor. The output feature map of the last 2D convolutional layer serves as the final output of the stacked 2D convolutional layers.

[0056] For the In the layer, Output channels at position The eigenvalues ​​of are defined as follows:

[0057]

[0058] in, For the Tier convolution kernel weights, where is the index of the convolution kernel in the height and width directions, is the index of the input channel, is the index of the output channel, and the kernel size is ; Output channel Represents the first feature map generated by the convolution kernel channels; The input feature map Tier Channels at position The value of is the bias value; is the ReLU activation function; the number of channels of the input tensor is , the number of output tensor channels is determined by the number of set convolution kernels;

[0059] In this process, it is necessary to add a pooling layer after each two-dimensional convolution layer to downsample and output;

[0060] S404. Process the input tensor through multiple stacked one-dimensional convolutional layers; the input feature map of the first one-dimensional convolutional layer is the input tensor, the output feature map of the previous one-dimensional convolutional layer is used as the input feature map of the next one-dimensional convolutional layer, and the output of the last one-dimensional convolutional layer is used as the final output of the multi-stacked one-dimensional convolutional layer;

[0061] The multi-layer stacked one-dimensional convolutional layer and the multi-layer stacked two-dimensional convolutional layer are parallel different branch structures;

[0062] In this process, it is necessary to add a pooling layer after each one-dimensional convolution layer to downsample and output;

[0063] S405. By merging the outputs of different branch structures, that is, splicing the outputs of multiple stacked 2D convolutional layers and multiple stacked 1D convolutional layers, a spliced ​​feature map is obtained. The feature map is then flattened and the extracted features are integrated through a fully connected network to finally output the fault status, cumulative health index, and remaining life prediction. is a multidimensional vector:

[0064] .

[0065] Model training and optimization:

[0066] Loss function design

[0067] The optimization design of the CNN model includes:

[0068] A1. Assume that the CNN model is updated using mini-batch gradient descent. Divide the samples in the training set into multiple batches, each containing N samples. Update the CNN model using the samples in each batch:

[0069] For any batch of samples, based on the constructed CNN network, for multi-task output, the loss function is designed as a weighted sum of multiple sub-losses, including:

[0070] For each sample in the current batch, the sample features are used as the input of the CNN model, and the output of the CNN model is used as the predicted value;

[0071] For the prediction of fault status, cumulative health index and remaining life, the total loss Expressed as:

[0072]

[0073] , To adjust the importance weight of each part of the loss;

[0074] in, The loss function representing the fault state adopts the cross entropy loss function:

[0075]

[0076] Among them, in the current batch of samples, the health status label and predicted value of the qth sample are recorded as 、 ;q=1,2,…,Q;

[0077] Represents the cumulative health index loss function:

[0078]

[0079] Among them, in the current batch of samples, the cumulative health index label and predicted value of the qth sample are recorded as 、 ;q=1,2,…,Q;

[0080] Represents the remaining life loss function, using mean square error loss:

[0081]

[0082] Among them, in the current batch of samples, the cumulative remaining life label and predicted value of the qth sample are recorded as 、 ;q=1,2,…,Q;

[0083] A2. Repeat step A1 for each batch of samples in the training set, and then train the CNN model based on the total loss function until the model converges or all batches of samples are trained. A mature CNN model is obtained for prediction.

[0084] In the embodiment of the present application, regularization and overfitting prevention can also be performed.

[0085] To avoid overfitting of the model, Dropout and L2 regularization can be added during training:

[0086] Dropout layer: Insert the Dropout layer into the fully connected layer so that some nodes are “dropped” with a certain probability during the training process, thereby enhancing the generalization ability of the model. Assume that the Dropout probability is , the output is:

[0087]

[0088] in, is the input of the Dropout layer, is the output of the Dropout layer, represents the Hadamard product, is a Bernoulli distributed random vector, for vector The jth element in , obeys ; ;Right now The probability of taking 1 is , The probability of taking 0 is .

[0089] L2 regularization: Add a weight decay term to the loss function and change the weight parameter The L2 norm of is added to the loss function:

[0090]

[0091] in, is the regularization parameter.

[0092] The label definitions for the three tasks are as follows:

[0093] Fault Status Tags: For Tags : Indicates that the cross belt sorter has failed within the time window; , indicating that the cross-belt sorter has no faults within the time window. The construction method is to backtrack the fault log and If a fault occurs in the future, it is 1, otherwise it is 0. The loss function is defined as ;

[0094] Cumulative health index label: continuous value label, through the defined parsing model Get (pseudo label), the loss function is defined above , although the CNN model is trained with the traditional health function The output of is optimized as a pseudo-supervisory label, but its final output Not limited to Instead of using the expression form, it learns stronger state expression ability through data-driven approach, which can completely replace Realize real-time prediction of health scores (cumulative health index), thus avoiding the complexity and limitations of traditional models in parameter tuning and function modeling;

[0095] Remaining life label: continuous value label, through the defined parsing model Get (pseudo label), the loss function is defined above Although the CNN model is trained with the traditional life estimation function The output of is optimized as a pseudo-supervisory label, but its final output Not limited to Instead of using analytical expressions, the implicit mapping relationship between device status and life span is automatically learned in a data-driven way. In the actual operation stage, the CNN output Can be used directly to replace traditional The model predicts the remaining life, thus avoiding the The accuracy and stability decreased due to modeling difficulties and health score error transmission problems, and the real-time and generalization capabilities of life expectancy prediction were improved.

[0096] Cumulative health index output in CNN model It can be considered a quantitative indicator of the current status of the equipment. By combining historical data with real-time data, a dynamic monitoring curve of the health status can be developed. Using time series analysis methods (such as sliding average or Kalman filtering), the health score data is smoothed in real time. At the same time, the changing trends of the fault status and remaining life prediction are combined to construct an early warning level. Specifically defined as:

[0097]

[0098] in, , is the weight parameter, It is the preset life threshold. Different maintenance strategies can be set according to the value of the warning index.

[0099] In the embodiments of the present application: when the warning index is between 0.0 and 0.3, it indicates that the equipment is in a healthy state and can continue to operate normally, and only status monitoring needs to be performed according to the regular cycle; when the index is in the range of 0.3 to 0.6, it is a medium risk, and it is recommended to arrange on-site engineers to conduct inspections and strengthen data collection and analysis of key parts; when the index reaches 0.6 to 0.8, it indicates that the equipment has a large potential risk, and a partial shutdown inspection should be carried out, and corresponding maintenance or replacement plans should be prepared; if the warning index exceeds 0.8, it is a high-risk or critical state, and the equipment must be shut down immediately, and emergency repairs or component replacements should be arranged as a priority to prevent equipment failure from causing a systemic shutdown.

[0100] This grading strategy can be further optimized by combining historical fault data and actual operating experience to achieve more accurate and efficient intelligent maintenance scheduling.

[0101] In the embodiments of this application, to meet real-time requirements at industrial sites, CNN models can be lightweighted through techniques such as pruning, quantization, and knowledge distillation, ensuring high-speed and efficient operation on low-power embedded devices. The deployed model, within the established digital twin system for logistics transfer sites, can perform online predictions alongside data collection, providing real-time decision support information for equipment maintenance personnel.

Claims

1. An intelligent maintenance method for small item sorting machines in logistics transfer yards based on digital twins, characterized by: The following steps are involved: S1. Deploy multiple sensors at key locations on the cross-belt sorter to collect data in real time, transmit it to the preprocessing module for preprocessing, obtain preprocessed data vectors, and construct a data matrix within the time window; S2. Build a digital twin model of the cross-belt sorter; S3. Calculate the cumulative health index and remaining life of the cross-belt sorter based on historical data, and then construct a training set based on fault status, cumulative health index, and remaining life; S4. Build a CNN model to predict fault status, cumulative health index, and remaining life, optimize the CNN model, and calculate the early warning index to specify the maintenance strategy; In S4, a CNN model is constructed to perform fault diagnosis, cumulative health index, and remaining life prediction: S401. Input Data Tensor Construction: After the data preprocessing module, the real-time sensor data is normalized, denoised, and windowed to form the following data matrix: Reconstruct the data into a three-dimensional tensor format, set in the form of "height-width-channel", where the height , the m corresponds to the number of deployed sensors, the width , T corresponds to the time step, each sensor data constitutes a channel separately, Indicates the number of channels of the input tensor; the final input tensor is recorded as ; S402. Randomly crop the input tensor along the time axis, inject noise to simulate sensor errors, and randomly scale the input tensor to expand the dataset, allowing the model to better adapt to data variations in actual work. S403. By stacking multiple 2D convolutional layers, the model gradually captures features from local to global. In the stacked 2D convolutional layers, the output feature map of the previous 2D convolutional layer serves as the input feature map of the next 2D convolutional layer. The input feature map of the first 2D convolutional layer serves as the input tensor. The output feature map of the last 2D convolutional layer serves as the final output of the stacked 2D convolutional layers. For the In the layer, Output channels at position The eigenvalues ​​of are defined as follows: in, For the Tier convolution kernel weights, where is the index of the convolution kernel in the height and width directions, is the index of the input channel, is the index of the output channel, and the kernel size is ; Output channel Represents the first feature map generated by the convolution kernel channels; The input feature map Tier Channels at position The value of is the bias value; is the ReLU activation function; the number of channels of the input tensor is , the number of output tensor channels is determined by the number of set convolution kernels; Add a pooling layer after each two-dimensional convolution layer to downsample and output; S404. Process the input tensor through multiple stacked one-dimensional convolutional layers; the input feature map of the first one-dimensional convolutional layer is the input tensor, the output feature map of the previous one-dimensional convolutional layer is used as the input feature map of the next one-dimensional convolutional layer, and the output of the last one-dimensional convolutional layer is used as the final output of the multi-stacked one-dimensional convolutional layer; The multi-layer stacked one-dimensional convolutional layer and the multi-layer stacked two-dimensional convolutional layer are parallel different branch structures; Add a pooling layer after each one-dimensional convolution layer to downsample and output; S405. By merging the outputs of different branch structures, that is, splicing the outputs of multiple stacked 2D convolutional layers and multiple stacked 1D convolutional layers, a spliced ​​feature map is obtained. The feature map is then flattened and the extracted features are integrated through a fully connected network to finally output the fault status, cumulative health index, and remaining life prediction. is a multidimensional vector: ； They represent the predicted values ​​of fault status, cumulative health index and remaining life respectively.

2. The intelligent maintenance method for small item sorting machines in logistics transfer yards based on digital twins according to claim 1 is characterized by: The key positions of the cross-belt sorter include the conveyor belt, sorting grid and scanner; The various sensors are used to collect vibration, temperature, operating time, current and noise information.

3. The intelligent maintenance method for small item sorting machines in logistics transfer yards based on digital twins according to claim 1 is characterized by: The data of the various sensors are collected by a camera and then transmitted to a PLC controller, which then transmits the data to a data preprocessing module via industrial Ethernet or a field bus. The data preprocessing module processes the collected data including denoising and normalization.

4. The intelligent maintenance method for small item sorting machines in logistics transfer yards based on digital twins according to claim 1 is characterized by: The construction of the digital twin model of the cross-belt sorter includes: Based on the CAD model, kinematic and dynamic parameters of the actual cross-belt sorter, an accurate three-dimensional model of the cross-belt sorter is established, and the pre-processed data vectors are imported into the digital twin model to achieve synchronization between the digital twin model of the sorter and the actual data.

5. The intelligent maintenance method for small item sorting machines in logistics transfer yards based on digital twins according to claim 1 is characterized by: The S3 includes: S301. Obtain data in multiple time windows according to step S1. The data in each time window is recorded as: S302. For data within any time window, calculate the cumulative health index and estimate the remaining lifespan, including: The cumulative health index is obtained by integrating the time : in, For the mapping function: in, Represents the mapping function input variable, To adjust the parameters; It is used to reflect the nonlinear relationship between the cumulative health index and the overall status of the equipment, expressed as: in For the i The weight of each sensor feature; ; Based on the accumulated health status and the preset health threshold , estimated remaining life : in, is the equipment health degradation rate; S303. Take any time window data as sample feature and add cumulative health index , Remaining life and as a label; is the fault status label, : , indicating that the cross-belt sorter has experienced a fault within the time window; , indicating that the cross-belt sorter has no faults within the time window; is the cumulative health index label; It is the remaining life label; The training sample of time window data is recorded as ; S304. For each time window data, repeat steps S302 to S303 to obtain the training samples corresponding to each time window data, add them to the same set to form a training set.

6. The intelligent maintenance method for small item sorting machines in logistics transfer yards based on digital twins according to claim 1 is characterized by: The optimization design of the CNN model includes: A1. Assume that the CNN model is updated using mini-batch gradient descent. Divide the samples in the training set into multiple batches, each containing N samples. Update the CNN model using the samples in each batch: For any batch of samples, based on the constructed CNN network, for multi-task output, the loss function is designed as a weighted sum of multiple sub-losses, including: For each sample in the current batch, the sample features are used as the input of the CNN model, and the output of the CNN model is used as the predicted value; For the prediction of fault status, cumulative health index and remaining life, the total loss Expressed as: To adjust the importance weight of each part of the loss; in, The loss function representing the fault state adopts the cross entropy loss function: Among them, in the current batch of samples, the health status label and predicted value of the qth sample are recorded as 、 ;q=1,2,…,Q; Represents the cumulative health index loss function: Among them, in the current batch of samples, the cumulative health index label and predicted value of the qth sample are recorded as 、 ;q=1,2,…,Q; Represents the remaining life loss function, using mean square error loss: Among them, in the current batch of samples, the cumulative remaining life label and predicted value of the qth sample are recorded as 、 ;q=1,2,…,Q; A2. Repeat step A1 for each batch of samples in the training set, and then train the CNN model based on the total loss function until the model converges or all batches of samples are trained. A mature CNN model is obtained for prediction.

7. The intelligent maintenance method for small item sorting machines in logistics transfer yards based on digital twins according to claim 1 is characterized by: The calculation of the early warning index and the designation of the maintenance strategy include: Cumulative health index output in CNN model It is considered as a quantitative indicator of the current status of the equipment. The early warning level is constructed by combining the changing trend of the fault status and the remaining life prediction. The specific definition is: in, is the weight parameter, is the preset lifespan threshold; Different maintenance strategies are set according to the value of the warning index. Specifically, the warning index is divided into intervals, and different maintenance strategies are formulated for each interval warning index.

Images