Logistics equipment fault prediction method and system based on digital twin neural network

Abstract

The invention provides a logistics equipment fault prediction method and system based on a digital twin neural network, and the method comprises the following steps: S1, collecting target data which at least comprises the parameters of a battery, a motor and a navigation system; s2, constructing a digital twin model of the AGV according to the target data, simulating a fault condition when the AGV runs, recording data under the fault condition, and generating a simulation data set; s3, constructing a neural network model according to the simulation data set to predict the state change of the AGV; s4, constructing an AGV monitoring platform, integrating the neural network model, performing fault prediction on the AGV, and outputting a prediction result; s5, setting an early warning mechanism, analyzing the prediction result, and triggering early warning when a potential fault is predicted; by combining the digital twin model and the neural network model, the accuracy of fault prediction can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of large language models, and in particular to a logistics equipment fault prediction method and system based on digital twin neural network. Background Art

[0002] In modern logistics systems, automated guided vehicles (AGVs), as one of the most important logistics equipment, are widely used in warehousing, manufacturing and other scenarios. The operating status of AGVs directly affects the efficiency and reliability of the entire logistics system. However, AGVs are prone to various faults during long-term operation, such as insufficient battery power, motor overheating, navigation system abnormalities, etc. If these faults are not discovered and handled in time, they will lead to equipment downtime, logistics system interruption, and even safety accidents.

[0003] In the existing technology, AGV fault prediction and maintenance mainly rely on regular inspections and passive maintenance. Regular inspections and passive maintenance are difficult to accurately predict the failure time and type of equipment, which can easily lead to failure to detect faults in time and affect the normal operation of the equipment. In addition, AGV fault prediction methods mostly rely on a single data source and cannot fully and accurately reflect the operating status of the equipment, making it difficult to achieve efficient fault prediction and maintenance. Summary of the invention

[0004] The purpose of this application is to provide a logistics equipment fault prediction method and system based on digital twin neural network, build a comprehensive data set through the digital twin model, and use the neural network model to predict faults, so as to improve the accuracy and reliability of the prediction results.

[0005] In order to achieve the above-mentioned objectives, one aspect of the present application provides a logistics equipment fault prediction method based on a digital twin neural network, comprising the following steps: S1: collecting target data, wherein the target data includes at least parameters of the battery, motor and navigation system; S2: constructing a digital twin model, simulating fault conditions and recording the data under the fault conditions to generate a simulated data set; S3: constructing a neural network model to predict the state changes of the AGV; S4: constructing a monitoring platform and integrating the neural network model to perform fault prediction; S5: setting up an early warning mechanism to trigger an early warning when a potential fault is predicted.

[0006] Preferably, the parameters of the battery, motor and navigation system include: battery voltage, current and temperature; motor speed, torque and vibration; and navigation system position, speed and direction, respectively.

[0007] Preferably, building a digital twin model includes the following steps: S201: geometric modeling, using modeling software to create a three-dimensional model of the AGV and importing it into the simulation software; S202: physical modeling, defining the physical properties of the AGV in the simulation software; S203: sensor modeling, adding a sensor model corresponding to the AGV in the simulation software; S204: control system modeling, simulating the control system of the AGV, including navigation algorithms, path planning algorithms and obstacle avoidance algorithms.

[0008] Preferably, S2 further includes: setting a number of fault scenarios in the simulation software, recording data of normal operation of the AGV, and the simulation data set also includes data of normal operation of the AGV.

[0009] Preferably, constructing the neural network model includes the following steps: S301: obtaining a simulation data set of AGV; S302: using a part of the simulation data in the simulation data set to train the neural network model; S303: using another part of the simulation data in the simulation data set to verify the prediction results of the neural network model; S304: obtaining the simulation results of the digital twin model, and comparing the simulation results with the prediction results of the neural network model to evaluate the accuracy of the neural network model.

[0010] Preferably, setting the early warning mechanism includes: setting an early warning threshold, and triggering an early warning when the prediction result exceeds the early warning threshold.

[0011] Preferably, the target data also includes historical operation records and maintenance records, wherein the historical operation records include AGV operation paths, operation time and load information, and the maintenance records include maintenance history, fault logs and repair records.

[0012] Preferably, S301 further includes obtaining target data of the AGV;

[0013] Using a part of the simulated data in the simulated data set to train the neural network model includes: S3021: processing the target data to form a real data set; S3022: aligning and integrating a part of the real data in the real data set with a part of the simulated data in the simulated data set at the data level to form a training set; S3023: inputting the data in the training set into the neural network model for training;

[0014] Using another part of the simulated data in the simulated data set to verify the prediction results of the neural network model includes: S3031: aligning and integrating another part of the real data in the real data set with another part of the simulated data in the simulated data set at the data level to form a verification set: S3032: inputting the data in the verification set into the neural network model for verification.

[0015] On the other hand, the present application provides a logistics equipment fault prediction system based on a digital twin neural network, using any of the fault prediction methods described above, including: a data acquisition layer, including sensors and data acquisition units, configured to collect AGV data; a data processing layer, including a data processing unit and a data storage unit, configured to process, analyze and store data; a model prediction layer, including a digital twin model and a neural network model, configured to predict faults of AGVs; a user interface layer, including a monitoring platform and an early warning unit, configured to output prediction results and early warning information to users.

[0016] Preferably, the sensors include at least a battery monitoring sensor, a motor sensor and a navigation system sensor.

[0017] The technical solution provided by this application can achieve the following beneficial effects:

[0018] 1. This application can accurately simulate and predict the operating status of AGV equipment by combining digital twin models and neural network models. The digital twin model provides detailed simulation data, and the neural network model predicts future status by learning these data, thereby achieving early warning of potential failures. This method can effectively improve the accuracy of fault prediction, identify potential equipment problems in advance, and reduce equipment downtime and maintenance costs.

[0019] 2. The integrated comparison mechanism between real-time data and prediction results in the solution makes equipment status monitoring more comprehensive and dynamic. Real-time data provides feedback on the actual status of the current equipment, while the prediction results provide early warning of future status. Combining the two, the system's adaptability and response speed are enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0021] Figure 1 It is a flow chart of a logistics equipment fault prediction method based on a digital twin neural network provided in an embodiment of the present application;

[0022] Figure 2 It is a flow chart of the digital twin model construction method provided in the embodiment of the present application;

[0023] Figure 3 is a flow chart of a method for constructing a neural network model provided in an embodiment of the present application;

[0024] Figure 4 is a flow chart of a training method for a neural network model provided in an embodiment of the present application;

[0025] Figure 5 It is a flow chart of a verification method of a neural network model provided in an embodiment of the present application;

[0026] Figure 6 It is a structural schematic diagram of a logistics equipment fault prediction system based on a digital twin neural network provided in an embodiment of the present application;

[0027] Figure 7 It is a structural diagram of the data acquisition layer provided in an embodiment of the present application;

[0028] Figure 8 It is a structural diagram of the data processing layer provided in an embodiment of the present application;

[0029] Fig. 9 It is a schematic diagram of the structure of the model prediction layer provided in the embodiment of the present application;

[0030] Fig.10 is a structural diagram of the user interface layer provided in an embodiment of the present application;

[0031] Among them: 10-data acquisition layer, 101-data acquisition unit, 102-battery monitoring sensor, 103-motor sensor, 104-navigation system sensor, 20-data processing layer, 201-data processing unit, 202-data storage unit, 30-model prediction layer, 301-digital twin model, 302-neural network model, 40-user interface layer, 401-monitoring platform, 402-early warning unit. DETAILED DESCRIPTION

[0032] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0033] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0034] On the one hand, the present application provides a logistics equipment fault prediction method based on a digital twin neural network, such as Figure 1 As shown, the following steps S1-S5 are included:

[0035] S1: Collect target data, where the target data at least includes parameters of a battery, a motor, and a navigation system.

[0036] It can be imagined that battery parameters include battery voltage, current and temperature; motor parameters include motor speed, torque and vibration; navigation system parameters include navigation system position, speed and direction.

[0037] Specifically, the target data involves extracting key operating parameters and status information from the AGV, and the sources of the target data involve three major types of sensors, including battery monitoring sensors, motor sensors, and navigation system sensors.

[0038] The battery monitoring sensor is directly connected to the battery cell of the AGV to measure the voltage, current and temperature of the AGV battery cell. Furthermore, the battery monitoring sensor includes a battery voltage sensing module, a battery current sensing module, a battery temperature sensing module and a charging state sensing module. The battery voltage sensing module monitors the voltage of the AGV battery cell to determine whether the AGV battery cell is in a normal working range; the battery current sensing module monitors the charge and discharge current of the AGV battery cell to evaluate the load of the AGV battery cell; the battery temperature sensing module records the operating temperature of the AGV battery cell to ensure that it is within a safe range; the charging state sensing module tracks the charging and discharging status of the AGV battery cell and provides comprehensive data on the health status of the AGV battery cell.

[0039] The motor sensor is installed on the housing or motor shaft of the AGV motor. Further, the motor sensor includes a motor vibration sensor, a motor temperature sensor, a motor speed sensor and a motor current sensor. The motor vibration sensor monitors the vibration of the motor to detect whether there is abnormal vibration or failure; the motor temperature sensor records the operating temperature of the motor to prevent damage caused by overheating; the motor speed sensor measures the speed of the motor to ensure that it is within the normal operating range; the motor current sensor monitors the current consumption of the motor to evaluate its workload.

[0040] The navigation system sensor is installed in the navigation system part of the AGV, such as above the AGV body or near the wheels. Furthermore, the types of navigation system sensors include GPS sensors, laser rangefinders, inertial measurement units (IMUs) and gyroscopes. This application may select one or more of the above types. The GPS sensor records the location data of the AGV to track its movement trajectory; the laser rangefinder measures the distance between the AGV and surrounding obstacles to prevent collisions and ensure safe navigation; the inertial measurement unit obtains the AGV's acceleration, angular velocity and other dynamic data to evaluate its motion state; the gyroscope provides the AGV with angle information to ensure its navigation accuracy.

[0041] When using the above-mentioned battery monitoring sensor, motor sensor, and navigation system sensor, set the sampling frequency of the sensor to sample every second or more, and use wireless communication technology for data transmission, such as Wi-Fi, Bluetooth, or 4G / 5G network.

[0042] It is conceivable that a database is established to store the collected real-time data, ensure the integrity and traceability of the data, and regularly back up the data to prevent data loss or damage and ensure the reliability of the system. Optional databases include SQL databases and NoSQL databases, such as MySQL, PostgreSQL, and MongoDB databases.

[0043] Furthermore, the data from different sensors are unified to the same scale range, that is, data normalization, to ensure that the impact of each data on the model is within the same range. There are mainly two methods: minimum-maximum normalization, Z-Score normalization, and unit vector normalization.

[0044] In the minimum-maximum normalization method, the data is scaled to between [0,1], and the formula is as follows:

[0045]

[0046] Among them, x is the original data, x min is the minimum value in the data set, x max is the maximum value in the data set, x normis the standardized data.

[0047] In the Z-Score standardization method, the data is converted to a standard normal distribution with a mean of 0 and a standard deviation of 1. The formula is as follows:

[0048]

[0049] Among them, x is the original data, μ is the mean of the data, σ is the standard deviation of the data, and x std is the standardized data.

[0050] In the unit vector normalization method, the data is normalized to the unit vector length, and the formula is as follows:

[0051]

[0052] Here, ||x|| is the norm (L2 norm) of vector x.

[0053] It can be understood that if the range of the data is within a known interval (such as 0 to 100), the minimum-maximum normalization method is used; if the distribution of the data is close to a normal distribution, the use of Z-Score normalization can better handle outliers and deviations in the data; for feature vectors that need to be normalized to unit length, unit vector normalization is used.

[0054] In an example, a set of sensor data is collected, including battery voltage in the range of [10V, 15V], motor temperature in the range of [20℃, 80℃], and navigation system position in the range of [0m, 100m]. This example uses the minimum-maximum normalization method to process the battery voltage data. For the battery voltage data, the standardized value is calculated: For motor temperature data: For navigation system location:

[0055] By implementing standardization processing, data from different sensors can be unified to the same scale range, thereby improving the effectiveness of data analysis and model training.

[0056] In some examples, the target data includes not only real-time sensor data, but also historical operation records and maintenance records, wherein the historical operation records include AGV operation paths, operation times, and load information, and the maintenance records include maintenance history, fault logs, and repair records.

[0057] Specifically, the operation path data is used to record the specific movement trajectory of the AGV in the warehouse or manufacturing environment, including the starting point, end point and intermediate path points. The operation time data records the start time, end time and total operation time of each operation. The load information records the type, weight and volume of goods carried by the AGV each time it runs. The maintenance history records the regular maintenance of the equipment, including maintenance time, maintenance content and maintenance personnel. These data can be used to track the maintenance status of the equipment. The fault log records the fault events that occur during the operation of the AGV, including fault time, fault type and fault cause analysis. The maintenance record records the maintenance status after each fault, including maintenance time, maintenance content, replaced parts and maintenance personnel.

[0058] S2: Build a digital twin model according to the target data, simulate the fault condition during the operation of the AGV and record the data under the fault condition to generate a simulation data set.

[0059] Specifically, building a digital twin model can virtualize the behavior and characteristics of actual equipment and test various fault conditions in a simulated environment to generate a simulated data set. This process includes three main parts: modeling, simulation, and data generation.

[0060] For AGV equipment that actually needs to perform fault prediction, the actual operation data, technical specifications, and design parameters of the AGV are integrated to create an accurate digital twin model.

[0061] like Figure 2 As shown, building a digital twin model includes the following steps S201-S204:

[0062] S201: Geometric modeling, using CAD software to create a three-dimensional model of the AGV and importing it into Simulink software.

[0063] Furthermore, all key components of the AGV, such as the body, battery, motor, and navigation system, are identified, and the geometric dimensions and shape information of all components are obtained. These data come from the design drawings of the AGV, and a three-dimensional model of the AGV and its components is created using CAD software. Each component is modeled in detail to ensure that all important geometric features, such as connection points, installation locations, and structural details, are included, and the geometric models of each component are integrated into a complete AGV model. Optional modeling software also includes SolidWorks, which can be used to model the body, wheels, motors, and other components of the AGV in three dimensions. Optional simulation software also includes Ansys.

[0064] S202: Physical modeling, defining the physical characteristics of the AGV in Simulink.

[0065] Going further, physical properties include material properties and physical parameters, where the material properties of each component are defined, such as density, elastic modulus, and thermal conductivity, which affect the mechanical behavior and thermal characteristics of the model, and physical parameters are set, such as battery capacity, motor power, and transmission efficiency, to ensure that the model can accurately simulate the actual performance of the AGV.

[0066] S203: Sensor modeling: adding a sensor model corresponding to the AGV in the simulation software.

[0067] Furthermore, the installation position of the sensor on the AGV is determined, and the position and data interface of each sensor are configured in the model. The characteristic parameters of the sensor, such as the measurement range, resolution and sampling frequency, are defined. The output data of the sensor is simulated in the digital twin model, including the noise and error characteristics of the data, to make the simulation results more realistic.

[0068] S204: Control system modeling, simulating the control system of AGV, including navigation algorithm, path planning algorithm and obstacle avoidance algorithm.

[0069] Furthermore, the control strategy of AGV is defined, such as PID control, fuzzy control or adaptive control, and the mathematical model of the control algorithm is established, i.e., PID control, fuzzy control or adaptive control. The interface of the control system is simulated in the digital twin model, including input signals, control instructions and responses. The navigation algorithm, path planning algorithm and obstacle avoidance algorithm adopt the existing algorithms of AGV.

[0070] Fault simulation using a digital twin model includes the following steps:

[0071] S211: First define the fault types that need to be simulated, including battery overheating, motor failure and navigation system failure, and set the conditions and situations for the occurrence of the fault, such as battery temperature exceeding the standard, motor overload, etc.

[0072] S212: Inject fault conditions into the digital twin model to simulate different fault scenarios. This can be achieved by modifying input parameters, adjusting system behavior, and setting the time step, simulation duration, and data acquisition frequency in the model process.

[0073] S213: Start the digital twin model to simulate fault scenarios and collect simulation data in real time. The simulation data includes status information and fault characteristic data of various components such as batteries, motors, and navigation systems, and records the system's response and data under different fault conditions.

[0074] S214: Organize the recorded data into a structured format, such as CSV, a database, or a data table.

[0075] Furthermore, a label is added to each data sample. The label is used to identify the fault type and status of the data. The label may include information such as fault category, severity, and occurrence conditions. In addition, the simulation data can be classified to ensure that each type of fault has a sufficient sample size for subsequent model training and verification, and the sorted simulation data is stored in the database.

[0076] It is conceivable that in the fault scenario simulation, the AGV includes a normal operating state and a fault state, and the normal operating data and the fault data of the AGV are recorded at the same time, and the normal operating data and the fault data are integrated into a simulation data set.

[0077] S3: Construct a neural network model based on the simulated data set to predict the state change of the AGV.

[0078] Specifically, the neural network model is able to learn complex patterns from the data and is used to predict the status and failure conditions of the AGV.

[0079] like Figure 3 As shown, building a neural network model includes the following steps S301-S304:

[0080] S301: Obtain a simulation data set of AGV.

[0081] Furthermore, it also includes obtaining target data of the AGV, the target data including parameters of the battery, motor and navigation system, and collecting actual fault records of the AGV. The target data is cleaned to obtain a real data set.

[0082] The simulation data set is obtained from the data generated by the digital twin model. The simulation data set includes the normal operation data of the AGV and the data under various fault conditions.

[0083] The real data set is divided into a training set and a validation set, and the simulated data set is also divided into a training set and a validation set. The training set of the real data set and the training set of the simulated data set are combined into the final training set, and the validation set of the real data set and the validation set of the simulated data set are combined into the final validation set.

[0084] S302: Use a portion of the simulated data in the simulated data set to train the neural network model.

[0085] Furthermore, neural network models include feedforward neural networks, convolutional neural networks, recurrent neural networks and long short-term memory networks. Among them, feedforward neural networks are used for basic prediction tasks and are suitable for simple relationship modeling; convolutional neural networks are used to process data with spatial structures, such as image data, and can be used to extract local features in the data; recurrent neural networks are used to process time series data and can capture the time dependency in the data; long short-term memory networks are used to deal with long-term dependency problems and are suitable for time series prediction.

[0086] This example selects the long short-term memory network (LSTM). You can use a single-layer or multi-layer network structure to design the number of layers of the long short-term memory network, including an input layer, one or more LSTM layers, and an output layer. Specifically, the input layer is designed to accept time series data, and the input shape is (time step, number of features). For example, if the time step is 10 and the number of features is 5, the input layer shape is (10,5). A single-layer LSTM layer is used for simple time series prediction tasks. A multi-layer LSTM layer uses multiple LSTM layers stacked to capture more complex time dependencies. The output of each LSTM layer is used as the input of the next layer. In the LSTM layer, the number of units in the LSTM layer is selected, such as 50 or 100 units. The activation functions inside the LSTM layer usually include Sigmoid function and Tanh function. The Sigmoid function is used for the gating mechanism to control the inflow and outflow of information. Its calculation formula is as follows:

[0087]

[0088] Here, x is the input value and e is a natural constant, approximately equal to 2.71828.

[0089] The Tanh function is used to generate candidate values ​​for the cell state, and its calculation formula is as follows:

[0090]

[0091] Design the output layer to provide prediction results, using a linear activation function or a softmax activation function. In the softmax activation function, the output value is converted into a probability distribution, which is calculated as follows:

[0092]

[0093] Among them, x i is the i-th element of the output vector.

[0094] In model training, the loss function is defined and the mean square error (MSE) is used as the loss function. The calculation formula is as follows:

[0095]

[0096] Among them, y i is the true value, is the predicted value and N is the number of samples.

[0097] like Figure 4 As shown, the training method is as follows: S3021: processing the target data to form a real data set; S3022: aligning and integrating a part of the real data in the real data set with a part of the simulated data in the simulated data set at the data level to form a training set; S3023: inputting the data in the training set into the neural network model for training.

[0098] Furthermore, the Adam optimizer is used to minimize the loss function. Adam combines the momentum method (Momentum) and RMSprop, and adaptively adjusts the learning rate of each parameter. In the hyperparameter adjustment, the learning rate is set to 0.001, but it can be adjusted according to the model performance. The batch size can be selected as 32, 64, 128, etc. According to the scale of training data and computing resources, the number of training rounds is set, for example, 50 to 100 rounds. The specific number is determined according to the convergence of the model during training.

[0099] Use the planned training set to train the model and apply the Adam optimization algorithm to update the model parameters.

[0100] S303: Use another part of the simulated data in the simulated data set to verify the prediction result of the neural network model.

[0101] Furthermore, the model performance is evaluated using the planned validation set, and the hyperparameters are adjusted based on the validation results to optimize the model.

[0102] like Figure 5 As shown, the verification method is as follows: S3031: align and integrate another part of the real data in the real data set with another part of the simulated data in the simulated data set at the data level to form a verification set: S3032: input the data in the verification set into the neural network model for verification.

[0103] S304: Obtain simulation results of the digital twin model, and compare the simulation results with the prediction results of the neural network model to evaluate the accuracy of the neural network model.

[0104] Furthermore, the simulation results, i.e., simulation data, specifically include the device parameters of the AGV in various states, such as battery charge changes, motor temperature changes, navigation system status, etc. The simulation results are usually output in the form of time series data, including timestamps and corresponding device status parameters.

[0105] Calculate the error between the simulation results and the prediction results to evaluate the accuracy of the neural network model. The error calculation formula can choose the mean square error MSE or the mean absolute error MAE. According to the error analysis results, evaluate the accuracy and reliability of the model.

[0106] S4: Build an AGV monitoring platform and integrate the neural network model to predict AGV failures.

[0107] Furthermore, the overall architecture of the monitoring platform is designed, including the front-end, back-end and middle layer. The front-end provides an intuitive interface for operators to monitor and manage AGV equipment. The back-end includes data storage, processing and analysis modules, which are responsible for receiving, storing and processing sensor data. The middle layer realizes data transmission and processing and connects the front-end and back-end.

[0108] The monitoring platform can collect data from various sensors on AGV equipment in real time. The data types include battery monitoring data, motor operation data, navigation system data, etc., and store sensor data and prediction results.

[0109] The integrated neural network model includes model deployment and model integration. The trained neural network model is deployed as a service to provide a prediction interface. The deployment method is, for example, Docker container deployment, using Kubernetes for container orchestration. The prediction interface is designed as a RESTful API or gRPC interface for the monitoring platform to call for prediction; the monitoring platform mobilizes the prediction interface of the neural network model, transmits the real-time collected sensor data, and obtains the prediction results.

[0110] The monitoring platform displays the operating status and sensor data of AGV equipment in real time, and displays sensor data and prediction results in the form of charts, dashboards, etc.

[0111] Monitor sensor data in real time. When an abnormal situation is detected, trigger the early warning mechanism, input the collected sensor data in real time, call the neural network model for prediction, and send an alarm notification when the prediction results show that the equipment is at risk of failure. Provide corresponding maintenance suggestions based on the fault prediction results to help operators take timely measures.

[0112] S5: Setting up an early warning mechanism, analyzing the prediction results, and triggering an early warning when a potential failure is predicted.

[0113] Specifically, this example adopts a multi-level warning strategy, sets warnings of different levels, and grades them according to the severity of the fault. For example, three warning levels, low-level warning, medium-level warning, and high-level warning, are set, corresponding to minor faults, moderate faults, and serious faults of AGV equipment, respectively. Accordingly, the triggering conditions of each warning level are based on the specific threshold of the prediction results.

[0114] The setting of warning thresholds is based on historical data and expert experience, and the warning thresholds for various types of faults are defined. The threshold types include low battery power, high motor temperature, navigation system abnormalities, etc. It can be imagined that the warning thresholds can be dynamically adjusted according to the actual operating conditions and fault prediction effects.

[0115] Set warning rules and define warning trigger conditions for different fault types, for example, battery power is lower than a certain percentage, temperature exceeds a certain temperature value, navigation system failure rate exceeds a certain proportion, etc.

[0116] One early warning method is to collect sensor data and prediction results of the neural network model in real time, compare the sensor data and prediction results with the early warning threshold, determine whether to trigger an early warning, and trigger the corresponding early warning according to different early warning levels. Among them, the comparison of real-time sensor data is used to detect whether there is an abnormal situation at present, and the comparison of the prediction results with the early warning threshold is used to predict potential failures in advance.

[0117] Furthermore, multiple early warning notification methods are designed to ensure that early warning information is delivered to relevant personnel in a timely manner. Notification channels include SMS, email, system notification and mobile phone push, etc. The notification content includes fault type, warning level, equipment information and recommended measures.

[0118] For early warning notifications, this example provides two early warning processing methods, including automatic response and manual processing. Automatic response automatically triggers preset response measures based on the warning level and fault type. Response measures include reducing equipment load, adjusting operating parameters, and switching to backup systems. Manual processing notifies operators of early warning information for manual processing and decision-making. The system provides operation guides and maintenance suggestions to help operators respond quickly.

[0119] All warning events and handling situations are subsequently recorded to generate a warning log, in which the log content includes warning time, fault type, warning level and handling measures.

[0120] like Figure 6 As shown, this example provides a logistics equipment fault prediction system based on a digital twin neural network, using the above-mentioned fault prediction method, including a data acquisition layer, a data processing layer, a model prediction layer and a user interface layer, wherein the data acquisition layer includes sensors and data acquisition units, configured to collect AGV data; the data processing layer is configured to process and analyze the data; the model prediction layer includes a digital twin model and a neural network model, configured to predict faults of the AGV; the user interface layer includes a monitoring platform and an early warning unit, configured to output prediction results and early warning information to the user.

[0121] Specifically, Figure 7As shown in the figure, the configuration of the data acquisition layer includes two parts: sensor installation and data acquisition. Various sensors are installed on the AGV equipment, including battery monitoring sensors, motor temperature sensors, navigation system sensors, etc. It is conceivable that environmental sensors such as temperature, humidity, and vibration are installed in the AGV operating environment to monitor the impact of environmental conditions on equipment operation. The data acquisition unit realizes real-time data acquisition, historical data acquisition, and environmental data acquisition.

[0122] like Figure 8 As shown, the configuration of the data processing layer includes a data processing unit and a data storage unit. The data processing unit realizes data preprocessing and multi-source data fusion. Data preprocessing includes data cleaning, data normalization and data storage. Multi-source data fusion includes data integration and data fusion. Data integration integrates real-time data, historical data and environmental data to form a comprehensive equipment status data set, and then uses multi-source data fusion technology to comprehensively process data from different sources.

[0123] like Fig. 9 As shown, the configuration of the model prediction layer includes a digital twin model and a neural network model. The digital twin model simulates the working status of the AGV equipment and provides a training set and a validation set for the neural network model. The long short-term memory network model is selected as the neural network model. The long short-term memory network model predicts the future status of the AGV equipment and determines the potential failure risk.

[0124] like Fig.10 As shown in the figure, the configuration of the user interface layer includes a monitoring platform, an early warning unit, and user interaction. The monitoring platform provides real-time status monitoring of the AGV equipment, including battery power, motor temperature, navigation system status, etc., and displays the prediction results of the neural network model. When a potential fault is predicted, an early warning notification is triggered; the early warning unit promptly notifies relevant personnel of the early warning information through text messages, emails, system notifications, etc., and records all early warning events and processing situations, and generates an early warning log for subsequent analysis and improvement.

[0125] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A logistics equipment fault prediction method based on digital twin neural network, characterized in that: The following steps are involved: S1: Collect target data, where the target data at least includes parameters of a battery, a motor, and a navigation system; S2: constructing a digital twin model of the AGV according to the target data, simulating a fault condition during the operation of the AGV and recording data under the fault condition to generate a simulation data set; S3: constructing a neural network model according to the simulated data set to predict the state change of the AGV; S4: Build an AGV monitoring platform and integrate the neural network model to predict AGV failures and output prediction results; S5: Setting up an early warning mechanism, analyzing the prediction results, and triggering an early warning when a potential failure is predicted.

2. The method for predicting logistics equipment failure based on digital twin neural network according to claim 1 is characterized in that: The parameters of the battery, motor and navigation system are: Battery voltage, current and temperature; Motor speed, torque and vibration; and Navigation system position, speed and direction.

3. The method for predicting logistics equipment failure based on digital twin neural network according to claim 1 is characterized in that: Building a digital twin model involves the following steps: S201: Geometric modeling, using modeling software to create a three-dimensional model of the AGV and import it into the simulation software; S202: Physical modeling, defining the physical characteristics of the AGV in the simulation software; S203: sensor modeling, adding a sensor model corresponding to the AGV in the simulation software; S204: Control system modeling, simulating the control system of AGV, including navigation algorithm, path planning algorithm and obstacle avoidance algorithm.

4. The method for predicting logistics equipment failure based on digital twin neural network according to claim 1 is characterized in that: S2 also includes: Several fault scenarios are set in the simulation software, and the data of the normal operation of the AGV is recorded. The simulation data set also includes the data of the normal operation of the AGV.

5. The method for predicting logistics equipment failure based on digital twin neural network according to claim 4 is characterized in that: Building a neural network model involves the following steps: S301: Obtain a simulation data set of AGV; S302: training a neural network model using a portion of the simulated data in the simulated data set; S303: using another part of the simulated data in the simulated data set to verify the prediction result of the neural network model; S304: Obtain simulation results of the digital twin model, and compare the simulation results with the prediction results of the neural network model to evaluate the accuracy of the neural network model.

6. The method for predicting logistics equipment failure based on digital twin neural network according to claim 1 is characterized in that: Setting up early warning mechanisms includes: Set the warning threshold. When the prediction result exceeds the warning threshold, the warning is triggered.

7. The method for predicting logistics equipment failure based on digital twin neural network according to claim 5 is characterized in that: The target data also includes historical operation records and maintenance records, wherein the historical operation records include AGV operation paths, operation time and load information, and the maintenance records include maintenance history, fault logs and repair records.

8. The method for predicting logistics equipment failure based on digital twin neural network according to claim 7 is characterized in that: S301 also includes obtaining target data of the AGV; Using a portion of the simulated data in the simulated dataset to train the neural network model includes: S3021: Processing the target data to form a real data set; S3022: Aligning and integrating a portion of the real data in the real data set with a portion of the simulated data in the simulated data set at the data level to form a training set; S3023: inputting the data in the training set into a neural network model for training; Using another part of the simulated data in the simulated data set to verify the prediction results of the neural network model includes: S3031: Align and integrate another part of the real data in the real data set with another part of the simulated data in the simulated data set at the data level to form a verification set: S3032: Input the data in the verification set into the neural network model for verification.

9. A logistics equipment fault prediction system based on a digital twin neural network, using the generation method according to any one of claims 1 to 8, characterized in that: include: The data collection layer, including sensors and data collection units, is configured to collect AGV data; The data processing layer, including a data processing unit and a data storage unit, is configured to process, analyze and store data; The model prediction layer, including the digital twin model and the neural network model, is configured to predict the failure of the AGV; The user interface layer, including the monitoring platform and the early warning unit, is configured to output prediction results and early warning information to the user.

10. The logistics equipment fault prediction system based on digital twin neural network according to claim 9 is characterized in that: The sensors include at least a battery monitoring sensor, a motor sensor and a navigation system sensor.

Images