Intelligent three-dimensional warehouse full-process device coordination fault self-repair operation method

Abstract

The present application relates to the technical field of intelligent stereoscopic warehouse, and discloses a fault self-repairing operation method for full-process equipment cooperation of an intelligent stereoscopic warehouse, comprising the following steps: S1, collecting the running state data of the equipment in real time through the sensor groups arranged on each equipment in the stereoscopic warehouse; S2, sending the collected running state data to the edge computing node through the industrial Ethernet protocol; S3, analyzing the received running state data based on the pre-stored health state model by the edge computing node; the running state data of the full-process equipment of the stereoscopic warehouse is collected in real time through the deployment of the multi-source sensor groups, and the edge computing node realizes the online identification of the equipment health state and the accurate prediction of the fault occurrence time based on the long short-term memory network deep learning model, thereby overcoming the limitations that the traditional after-service maintenance and the single-machine predictive maintenance cannot cope with system cascade faults.

Description

Technical Field

[0001] This invention relates to the field of intelligent automated storage and retrieval systems (AS / RS) technology, specifically to a fault self-repair operation method for the entire process of intelligent automated storage and retrieval systems with coordinated equipment. Background Technology

[0002] With the deepening of Industry 4.0 and intelligent manufacturing, automated storage and retrieval systems (AS / RS) serve as the core hub of the modern logistics system. Their operational efficiency and stability directly determine the continuity of upstream production and downstream distribution. These highly dynamic systems are typically composed of a dense array of equipment, including stacker cranes, rail-guided shuttles (RGVs), conveyor lines, elevators, and depalletizing / palletizing robots. They accomplish high-density storage and throughput tasks through complex collaborative operations. However, under long-term, high-frequency operation, problems such as structural fatigue, electrical aging, and communication interference are unavoidable. Sudden failures in any link can trigger a chain reaction, leading to a break in the work chain and unplanned downtime, causing huge economic losses to enterprises.

[0003] To address this, we propose a fault self-healing operation method for the entire process of intelligent automated warehouse equipment collaboration. Summary of the Invention

[0004] The purpose of this invention is to provide a fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution, comprising the following steps: S1, real-time data collection of equipment operation status is achieved through sensor arrays deployed on various devices in the automated warehouse. The devices include stacker cranes, rail-guided shuttle vehicles (RGVs), conveyor lines, elevators, and depalletizing / unpalletizing robots. The operation status data includes vibration signals, current signals, temperature signals, image signals, and position encoding signals. S2 sends the collected operational status data to the edge computing node via the Industrial Ethernet protocol; S3, the edge computing node analyzes the received operating status data based on the pre-stored health status model, identifies the current health status of the device, and predicts the possible fault types and fault occurrence times; S4, When the health status model outputs an abnormal signal or an actual fault occurs, the edge computing node matches the corresponding self-repair strategy from the locally stored collaborative self-repair strategy library according to the fault type and severity. S5, the edge computing node transforms the matched self-healing strategy into specific control instructions. On the one hand, it sends the instructions directly to the controllers of the faulty equipment and its upstream and downstream equipment via the fieldbus. On the other hand, it sends the task adjustment requirements to the central scheduling system. The central scheduling system updates the global task queue according to the received task adjustment requirements and works with the edge computing node to execute the self-healing strategy. S6. After the self-repair strategy is completed, the edge computing node verifies the status of the faulty device and confirms whether the fault characteristics have disappeared and whether the device has returned to normal working status by comparing the operating status data before and after self-repair. S7 uploads all data generated during the self-healing process to the cloud server via wireless or wired network. The cloud server uses this data to incrementally train the health status model, update the model parameters, and distribute the updated model to the edge computing nodes.

[0006] Preferably, the sensor group includes a vibration sensor, a current sensor, a temperature sensor, a vision sensor, and an encoder; the vibration sensor is a triaxial accelerometer mounted on the bearing housings of the stacker crane's horizontal travel drive motor, lifting motor, fork extension motor, RGV travel wheel, and conveyor roller; the current sensor is a Hall effect closed-loop current sensor connected in series in the power supply lines of each drive motor.

[0007] Preferably, the specific construction process of the deep learning model based on the long short-term memory network is as follows: collect historical operating status data of the equipment under normal operation and various fault states, perform normalization preprocessing on the data, divide the data into time windows with a length of 10 seconds and a window sliding step of 1 second, use the data in each window as input features, use the remaining effective lifespan of the equipment as output label, use mean squared error as loss function, and train the network using the Adam optimizer. The network contains two LSTM layers with 128 neurons in each layer, followed by a fully connected layer.

[0008] Preferably, the collaborative self-healing strategy library contains multiple preset self-healing strategy program modules. Each module corresponds to a combination of fault type and severity level. The program modules are stored in the solid-state drive of the edge computing node in the form of executable code. The edge computing node calls the corresponding module by looking up the fault type-strategy mapping table. The fault type-strategy mapping table is a two-dimensional table, with the first dimension being the fault type and the second dimension being the severity level. Each cell in the table stores the entry address of the corresponding strategy module.

[0009] Preferably, for L1 level communication timeout faults, the matching self-repair strategy is a software reset and retry strategy, specifically including: after the edge computing node detects a communication timeout with a device controller or the device controller reports a momentary overcurrent alarm signal, it cuts off the power supply of the device through an output relay and maintains it for 500ms, then reconnects the power supply, sends a zero-return command to the device controller so that it moves 50mm in a safe direction at a low speed and then stops, then reissues the original task command and starts timing. If a task completion confirmation signal is received within 5 seconds, the self-repair is successful; otherwise, it is judged as a repair failure and upgraded to an L2 level fault. At the same time, the edge computing node sends a task waiting signal to the central scheduling system, and the central scheduling system suspends issuing new tasks to the device and its direct upstream devices and suspends subsequent tasks; for L1 level positioning deviation faults, the matching self-repair strategy is a visually guided dynamic correction strategy.

[0010] Preferably, the fieldbus adopts the EtherCAT protocol, with the edge computing node acting as the EtherCAT master and each device servo driver acting as the EtherCAT slave. They are connected through a ring or linear topology, with a periodic synchronization time of 1ms. The edge computing node encapsulates the self-healing instructions into EtherCAT process data objects and sends them to the designated slave in each communication cycle, while simultaneously receiving status data fed back by the slave.

[0011] Preferably, the verification of the status of the faulty equipment specifically includes: confirming through encoder feedback signals that the moving parts have returned to a safe position and the position error is less than 1mm; acquiring images of key parts through a vision sensor and using an image recognition algorithm to confirm that the cargo posture is normal, without tilting or falling off.

[0012] Preferably, the incremental training adopts an online learning method. The cloud server merges the newly received self-repair process data with historical data to form a new training set. The long short-term memory network model is fine-tuned using the mini-batch gradient descent method. The learning rate is set to 0.001, the batch size is 32, and the training rounds are 10. After training is completed, the updated model is distributed to the edge computing node via the MQTT protocol. The edge computing node loads the new model and replaces the old model.

[0013] Preferably, the method also includes the step of constructing a digital twin model of the automated warehouse: using 3D modeling software to create geometric models of all equipment in the automated warehouse, including stacker cranes, shelves, conveyor lines, RGVs and goods, and assigning kinematic constraints and physical properties to each component; importing the geometric model into the Unity3D engine to create a virtual scene; connecting edge computing nodes through the OPC UA protocol to obtain real-time equipment operating status data and mapping it onto the virtual model to drive the virtual model to move synchronously.

[0014] Preferably, it also includes a sensor group deployed on each device in the automated warehouse for real-time acquisition of equipment operating status data. The sensor group includes a vibration sensor, a current sensor, a temperature sensor, a vision sensor, and an encoder. The vibration sensor is a PCB Piezotronics triaxial ICP accelerometer, the current sensor is a LEM Hall effect closed-loop current sensor, the temperature sensor is a PT100 platinum resistance thermometer, the vision sensor is a Basler CMOS industrial camera equipped with a fixed-focus lens and a ring LED fill light, and the encoder is a Heidenhain incremental encoder.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention utilizes a multi-source sensor array to collect real-time operational status data of all equipment in an automated warehouse. Edge computing nodes, based on a long short-term memory network deep learning model, enable online identification of equipment health status and accurate prediction of fault occurrence times, overcoming the limitations of traditional reactive maintenance and single-machine predictive maintenance in handling cascading system failures. When an anomaly or fault is triggered, the edge computing nodes match the corresponding self-healing program module from a collaborative self-healing strategy library based on the fault type and severity. On one hand, control commands are sent to the controllers of the faulty equipment and its upstream and downstream devices via the EtherCAT high-speed fieldbus, achieving dynamic correction, load balancing, or redundant takeover through multi-machine collaboration. On the other hand, any... The system synchronizes task adjustment requirements to the central dispatch system to update the global task queue, ensuring that fault self-healing and adaptive adjustment of logistics rhythm are completed without downtime, thereby significantly reducing unplanned downtime and the frequency of manual intervention. After the self-repair is completed, the system verifies through multi-sensor fusion to ensure that the equipment returns to normal working condition, and uploads the data of the entire self-repair process to the cloud server for incremental training, continuously optimizing the accuracy and adaptability of the health status model, enabling the system to learn and evolve from each fault, and finally achieving an intelligent leap from "passive response" to "proactive prevention-collaborative self-healing-closed-loop evolution", which greatly improves the reliability, continuity and robustness of intelligent automated warehouse operation. Attached Figure Description

[0016] Figure 1 This is a flowchart of a fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The present invention provides a technical solution comprising the following steps: S1 collects real-time operational status data of the equipment through sensor groups deployed on various devices in the automated warehouse. The equipment includes stacker cranes, rail shuttle vehicles (RGVs), conveyor lines, elevators, and depalletizing / palletizing robots. The operational status data includes vibration signals, current signals, temperature signals, image signals, and position encoding signals. S2 sends the collected operational status data to the edge computing node via the industrial Ethernet protocol. The edge computing node is an embedded computer deployed on-site in the automated warehouse, running a real-time operating system and having a built-in health status analysis engine. S3, the edge computing node analyzes the received operating status data based on the pre-stored health status model, identifies the current health status of the device and predicts the possible fault types and fault occurrence times. The health status model is a deep learning model based on a long short-term memory network. Its input is the time series of historical operating status data and its output is the remaining effective lifespan of the device. At the same time, a dynamic threshold algorithm is used to calculate the threshold range of health indicators in real time based on the current load, operating speed and ambient temperature of the device. S4. When the health status model outputs an abnormality or an actual fault occurs and a signal is triggered, the edge computing node matches the corresponding self-repair strategy from the locally stored collaborative self-repair strategy library according to the fault type and severity. The fault types include communication timeout fault, positioning deviation fault, drive motor overload fault, bearing wear fault and complete shutdown fault. The severity is divided into L1 level minor fault, L2 level medium fault and L3 level severe fault. S5, the edge computing node transforms the matched self-healing strategy into specific control instructions. On the one hand, it sends the instructions directly to the controllers of the faulty equipment and its upstream and downstream equipment via the fieldbus. On the other hand, it sends the task adjustment requirements to the central scheduling system. The central scheduling system updates the global task queue according to the received task adjustment requirements and works with the edge computing node to execute the self-healing strategy. S6. After the self-repair strategy is completed, the edge computing node verifies the status of the faulty device and confirms whether the fault characteristics have disappeared and whether the device has returned to normal working status by comparing the operating status data before and after self-repair. S7 uploads all data generated during the self-healing process to the cloud server via wireless or wired network. The cloud server uses this data to incrementally train the health status model, update the model parameters, and distribute the updated model to the edge computing nodes.

[0019] Furthermore, the sensor group includes vibration sensors, current sensors, temperature sensors, vision sensors, and encoders. The vibration sensors are triaxial accelerometers mounted on the bearing housings of the stacker crane's horizontal travel drive motor, lifting motor, fork extension motor, RGV travel wheel, and conveyor rollers, with a sampling frequency of at least 10kHz. The current sensors are Hall effect closed-loop current sensors connected in series in the power supply lines of each drive motor, with a sampling frequency of at least 5kHz. The temperature sensors are PT100 RTDs or thermocouples attached to the motor housing and bearing outer rings, with a sampling frequency of 1Hz. The vision sensors are CMOS industrial cameras mounted on the front of the stacker crane forks, at conveyor bends, and at the elevator entrance / exit, with a resolution of 1920×1080 pixels and a frame rate of 30fps. The encoders are photoelectric incremental encoders or absolute encoders coaxially mounted on the tail of the drive motor or the side of the rollers on each motion axis, with a pulse count of at least 2500 lines / revolution. The industrial Ethernet protocol uses the Profinet protocol, with edge computing nodes acting as Profinet IO controllers, and each device controller acting as a Profinet controller. The IO devices are connected in a star topology network via switches, with a data exchange cycle of 1ms. The edge computing nodes use Intel Celeron processors or ARM architecture processors, with no less than 4GB of memory and no less than 64GB of storage space. They run a Linux-based real-time operating system and have a Docker containerized environment installed. The health status analysis engine runs in the form of a container.

[0020] Furthermore, the specific construction process of the deep learning model based on Long Short-Term Memory (LSTM) networks is as follows: Historical operating status data under normal operation and various fault states of the equipment are collected. The data is preprocessed by normalization. The data is divided into time windows, each with a length of 10 seconds and a window sliding step of 1 second. The data within each window is used as input features, and the remaining effective lifespan of the equipment is used as the output label. Mean squared error is used as the loss function. The network is trained using the Adam optimizer. The network contains two LSTM layers, each with 128 neurons, followed by a fully connected layer. After training, the model is converted into a lightweight format suitable for edge computing and stored in the non-volatile memory of the edge computing node. The dynamic threshold algorithm is as follows: For each health indicator, the distribution of historical normal operation data under different load, speed, and temperature conditions is statistically analyzed. A regression model between the indicator and load, speed, and temperature is established. Real-time load, speed, and temperature are input into the regression model to obtain the predicted value of the indicator. A dynamic threshold band is formed by floating several times the standard deviation above and below the predicted value. When the real-time indicator exceeds this threshold band, it is judged as abnormal. The standard deviation is updated in real time based on historical data.

[0021] Furthermore, the collaborative self-healing strategy library contains multiple preset self-healing strategy program modules. Each module corresponds to a combination of fault type and severity level. The program modules are stored in the solid-state drives of the edge computing nodes in the form of executable code. The edge computing nodes call the corresponding modules by looking up the fault type-policy mapping table. The fault type-policy mapping table is a two-dimensional table, with the first dimension being the fault type and the second dimension being the severity level. Each cell in the table stores the entry address of the corresponding policy module.

[0022] Furthermore, for L1 level communication timeout faults, the matched self-repair strategy is a software reset and retry strategy, specifically including: after the edge computing node detects a communication timeout with a device controller or the device controller reports a momentary overcurrent alarm signal, it cuts off the power supply to the device via an output relay and maintains this for 500ms, then reconnects the power supply, sends a zero-return command to the device controller to make it move 50mm in a safe direction at low speed and then stop, then reissues the original task command and starts timing. If a task completion confirmation signal is received within 5 seconds, the self-repair is successful; otherwise, it is judged as a repair failure and upgraded to an L2 level fault. At the same time, the edge computing node sends a task waiting signal to the central scheduling system. The central dispatch system suspends issuing new tasks to the device and its direct upstream equipment and suspends subsequent tasks. For L1-level positioning deviation faults, the matched self-repair strategy is a vision-guided dynamic correction strategy, specifically including: when the stacker crane forks reach the predetermined position of the target storage location, the vision sensor acquires an image of the current storage location; the edge computing node preprocesses the image and calculates the actual physical deviation values ​​ΔX and ΔY of the pallet hole center relative to the fork reference point using a template matching algorithm; the deviation values ​​are then sent to the stacker crane's fork servo controller via a high-speed fieldbus; the fork servo controller dynamically adjusts the fork extension speed and direction using a proportional-integral control algorithm; and the vision sensor continuously acquires images during the fork movement. The system uses image processing to form a closed-loop control until the deviation value is less than the allowable error threshold of 2mm, and then completes the picking and placing of goods. For L2-level drive motor overload faults, the matched self-repair strategy is a load balancing and cycle time adjustment strategy. Specifically, when the lifting motor of a stacker crane experiences an overload that causes the current to continuously exceed 80% of the rated current for more than 10 seconds, the self-diagnosis module updates the equipment's performance parameters to reduce the maximum acceleration limit by 30% and the maximum speed limit by 20%, and synchronizes this information to the central dispatch system in real time. The central dispatch system maintains the equipment performance table, calculates the estimated time based on the required load weight and path length when assigning new tasks, and selects appropriate equipment. For performance-limited... The equipment is prioritized for light-load tasks or more distant storage locations, while the release rhythm of the upstream conveyor line is adjusted. For L2 level bearing wear failure, the matching self-repair strategy is an early warning and maintenance scheduling strategy, which includes: the health status model of the edge computing node identifies the amplitude of the bearing characteristic frequency exceeding the dynamic threshold through vibration spectrum analysis, and determines it as an early bearing wear failure. The failure information and the predicted remaining life are sent to the central scheduling system. The central scheduling system marks the equipment as a planned maintenance status and automatically generates a maintenance work order. Before the maintenance time window arrives, the task allocation is automatically adjusted to reduce the workload of the equipment. When the maintenance time window arrives, the equipment is stopped and the maintenance personnel are prompted to replace the bearing through the human-machine interface.For L3 level complete shutdown failures, the corresponding self-healing strategy is a physical redundancy collaborative takeover strategy. Specifically, when an edge computing node detects a device failure, it marks the device as failed, reads the pre-stored automated warehouse logistics topology map, and calls a path planning algorithm to calculate alternative paths for all affected goods from the current node to the target node. If an alternative path exists, it sends the new path to the central scheduling system to update the task queue and instructs upstream equipment to transport the goods along the new path. If no alternative path exists, a relay mode is initiated, controlling the conveyor line upstream of the failed device to transport the goods to the nearest temporary storage area. Simultaneously, idle AGVs are dispatched or manual forklifts are notified to retrieve the goods from the temporary storage area and transport them to the downstream connection point of the failed device, where downstream equipment continues transport.

[0023] Furthermore, the fieldbus adopts the EtherCAT protocol, with edge computing nodes acting as EtherCAT masters and each device's servo driver acting as an EtherCAT slave. They are connected through a ring or linear topology, with a periodic synchronization time of 1ms. The edge computing nodes encapsulate self-healing instructions into EtherCAT process data objects and send them to the designated slave in each communication cycle, while simultaneously receiving status data from the slave. The central scheduling system is the Warehouse Control System (WCS) deployed on a server, which exchanges data with the edge computing nodes through an OPCUA client.

[0024] Further verification of the faulty equipment status includes: confirming via encoder feedback signals that moving parts have returned to a safe position with a positional error of less than 1mm; acquiring images of key parts via vision sensors and using image recognition algorithms to confirm that the cargo posture is normal, without tilting or falling off; monitoring the motor's no-load operating current via current sensors and confirming that its deviation from the normal historical value is less than 10%; acquiring vibration signals via vibration sensors and comparing whether the amplitude of the fault characteristic frequency has dropped below the dynamic threshold; comparing all verification data with the data before self-repair to generate a verification report. If all indicators are normal, the repair is considered successful; otherwise, the repair is considered unsuccessful and a manual intervention alarm is issued.

[0025] Furthermore, incremental training employs an online learning approach. The cloud server merges newly received self-healing process data with historical data to form a new training set. The long short-term memory network model is fine-tuned using mini-batch gradient descent with a learning rate of 0.001, a batch size of 32, and 10 training epochs. After training, the updated model is distributed to edge computing nodes via the MQTT protocol. The edge computing nodes load the new model and replace the old model. The cloud server is deployed on a cloud platform using virtual server instances running the Ubuntu operating system, with the TensorFlow framework, MySQL database, and Mosquitto MQTT Broker installed.

[0026] Furthermore, the process includes the steps of constructing a digital twin model of the automated warehouse: using 3D modeling software to create geometric models of all equipment in the automated warehouse, including stacker cranes, shelves, conveyor lines, RGVs, and goods, and assigning kinematic constraints and physical properties to each component; importing the geometric model into the Unity3D engine to create a virtual scene; connecting edge computing nodes via the OPC UA protocol to obtain real-time equipment operating status data and mapping it onto the virtual model to drive the virtual model's synchronous movement; and using this digital twin model to perform fault simulation, i.e., artificially injecting fault data to observe the model's response to verify the effectiveness of the self-healing strategy.

[0027] Furthermore, it also includes a sensor array deployed on various devices in the automated warehouse to collect real-time operational status data. The sensor array includes vibration sensors, current sensors, temperature sensors, vision sensors, and encoders. The vibration sensor is a PCB Piezotronics triaxial ICP accelerometer, the current sensor is a LEM Hall effect closed-loop current sensor, the temperature sensor is a PT100 platinum resistance thermometer, and the vision sensor is a Basler... The system includes a CMOS industrial camera with a fixed-focus lens and a ring LED fill light, and a Heidenhain incremental encoder. Edge computing nodes, connected to the sensor group via signal cables, receive operational status data. They have a built-in health status analysis engine and a collaborative self-healing strategy library, performing fault prediction and diagnosis based on the health status model and generating collaborative control commands by matching self-healing strategies according to fault type. A central scheduling system, connected to the edge computing nodes via Ethernet, receives task adjustment requests, updates the global task queue, and collaboratively executes self-healing strategies with the edge computing nodes. The central scheduling system is a warehouse control system (WCS) deployed on a server. Actuators, including servo drives, frequency converters, motors, and controllers for each device, are connected to the edge computing nodes via a fieldbus. They receive control commands from the edge computing nodes and execute self-healing actions. The fieldbus uses EtherCAT, with the edge computing nodes acting as master stations. Slave stations include B&R ACOPOS micro servo drives for stacker cranes, Lenze 8400 frequency converters for conveyor lines, Kollmorgen AKD servo drives for RGVs, and Beckhoff BK9050 servo drives. The distributed I / O terminal module is connected by an EtherCAT coupler; the cloud server communicates with the edge computing nodes via a wireless or wired network to receive self-healing process and result data, store historical data, train and update the health status model, and distribute the updated model to the edge computing nodes. The cloud server is deployed on Alibaba Cloud or AWS cloud platforms using virtual server instances, running the Ubuntu operating system, and equipped with the TensorFlow framework, MySQL database, and MosquittoMQTT Broker; the digital twin platform is deployed on the cloud server or local workstation, using the Unity3D engine to build a 3D visualization model of the automated storage and retrieval system, receiving real-time device status data pushed by the edge computing nodes via the WebSocket protocol to drive the virtual model's movement, and displaying device health status and fault warning information through the user interface, while also providing fault simulation functionality; the human-machine interface uses an industrial touch screen or ordinary PC, connected to the central dispatch system via Ethernet, to display the overall operating status of the automated storage and retrieval system, the fault alarm list, and self-healing records, and allows operators to manually trigger self-healing strategies or cancel automatic repair. The human-machine interface is developed using HTML5 and JavaScript and communicates with the central dispatch system via WebSocket;Edge computing nodes have redundancy, meaning two identical edge computing nodes are configured, one as the primary node and the other as the backup node. The primary and backup nodes are connected via a heartbeat line and periodically send heartbeat signals to each other. If the backup node does not receive a heartbeat from the primary node for three consecutive cycles, it automatically takes over the primary node's functions, including taking over the IP address and EtherCAT master role.

[0028] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0029] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration, characterized in that: Includes the following steps: S1, real-time data collection of equipment operation status is achieved through sensor arrays deployed on various devices in the automated warehouse. The devices include stacker cranes, rail-guided shuttle vehicles (RGVs), conveyor lines, elevators, and depalletizing / unpalletizing robots. The operation status data includes vibration signals, current signals, temperature signals, image signals, and position encoding signals. S2 sends the collected operational status data to the edge computing node via the Industrial Ethernet protocol; S3, the edge computing node analyzes the received operating status data based on the pre-stored health status model, identifies the current health status of the device, and predicts the possible fault types and fault occurrence times; S4, When the health status model outputs an abnormal signal or an actual fault occurs, the edge computing node matches the corresponding self-repair strategy from the locally stored collaborative self-repair strategy library according to the fault type and severity. S5, the edge computing node transforms the matched self-healing strategy into specific control instructions. On the one hand, it sends the instructions directly to the controllers of the faulty equipment and its upstream and downstream equipment via the fieldbus. On the other hand, it sends the task adjustment requirements to the central scheduling system. The central scheduling system updates the global task queue according to the received task adjustment requirements and works with the edge computing node to execute the self-healing strategy. S6. After the self-repair strategy is completed, the edge computing node verifies the status of the faulty device and confirms whether the fault characteristics have disappeared and whether the device has returned to normal working status by comparing the operating status data before and after self-repair. S7 uploads all data generated during the self-healing process to the cloud server via wireless or wired network. The cloud server uses this data to incrementally train the health status model, update the model parameters, and distribute the updated model to the edge computing nodes.

2. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: The sensor group includes a vibration sensor, a current sensor, a temperature sensor, a vision sensor, and an encoder. The vibration sensor uses a triaxial accelerometer and is installed in the bearing housings of the stacker crane's horizontal travel drive motor, lifting motor, fork extension motor, RGV travel wheel, and conveyor roller. The current sensor uses a Hall effect closed-loop current sensor connected in series in the power supply lines of each drive motor.

3. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: The specific construction process of the deep learning model based on long short-term memory network is as follows: collect historical operating status data of the equipment under normal operation and various fault states, perform normalization preprocessing on the data, divide the data into time windows with a length of 10 seconds and a window sliding step of 1 second, use the data in each window as input features, use the remaining effective lifespan of the equipment as output label, use mean squared error as loss function, and train the network using Adam optimizer. The network contains two LSTM layers with 128 neurons each, followed by a fully connected layer.

4. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: The collaborative self-healing strategy library contains multiple preset self-healing strategy program modules. Each module corresponds to a combination of fault type and severity level. The program modules are stored in the solid-state drives of the edge computing nodes in the form of executable code. The edge computing nodes call the corresponding modules by looking up the fault type-policy mapping table. The fault type-policy mapping table is a two-dimensional table, with the first dimension being the fault type and the second dimension being the severity level. Each cell in the table stores the entry address of the corresponding policy module.

5. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: For L1 level communication timeout faults, the matching self-repair strategy is a software reset and retry strategy, specifically including: after the edge computing node detects a communication timeout with a device controller or the device controller reports a momentary overcurrent alarm signal, it cuts off the power supply to the device through an output relay and holds for 500ms, then reconnects the power supply, sends a zero-return command to the device controller to make it move 50mm in a safe direction at low speed and then stop, then reissues the original task command and starts timing. If a task completion confirmation signal is received within 5 seconds, the self-repair is successful; otherwise, it is judged as a repair failure and upgraded to an L2 level fault. At the same time, the edge computing node sends a task waiting signal to the central scheduling system, and the central scheduling system suspends issuing new tasks to the device and its direct upstream devices and suspends subsequent tasks. For L1 level positioning deviation faults, the matching self-repair strategy is a visually guided dynamic correction strategy.

6. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: The fieldbus adopts the EtherCAT protocol, with the edge computing node acting as the EtherCAT master and each device servo driver acting as the EtherCAT slave. They are connected through a ring or linear topology, with a periodic synchronization time of 1ms. The edge computing node encapsulates self-healing instructions into EtherCAT process data objects and sends them to the designated slave in each communication cycle, while simultaneously receiving status data from the slave.

7. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: The verification of the status of the faulty equipment specifically includes: confirming through encoder feedback signals that the moving parts have returned to a safe position and the position error is less than 1mm; and using a visual sensor to collect images of key parts and an image recognition algorithm to confirm that the cargo posture is normal, without tilting or falling off.

8. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: The incremental training adopts an online learning method. The cloud server merges the newly received self-repair process data with historical data to form a new training set. The long short-term memory network model is fine-tuned using the mini-batch gradient descent method. The learning rate is set to 0.001, the batch size is 32, and the training rounds are 10. After training is completed, the updated model is distributed to the edge computing nodes via the MQTT protocol. The edge computing nodes load the new model and replace the old model.

9. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: It also includes the steps of building a digital twin model of the automated warehouse: using 3D modeling software to create geometric models of all equipment in the automated warehouse, including stacker cranes, shelves, conveyor lines, RGVs and goods, and assigning kinematic constraints and physical properties to each component; importing the geometric model into the Unity3D engine to create a virtual scene; connecting edge computing nodes through the OPC UA protocol to obtain real-time equipment operating status data and mapping it onto the virtual model to drive the virtual model to move synchronously.

10. The fault self-repair operation method for the entire process of intelligent automated warehouse equipment collaboration according to claim 1, characterized in that: It also includes a sensor group, deployed on various devices in the automated warehouse to collect real-time data on the operating status of the equipment. The sensor group includes a vibration sensor, a current sensor, a temperature sensor, a vision sensor, and an encoder. The vibration sensor is a PCB Piezotronics triaxial ICP accelerometer, the current sensor is a LEM Hall effect closed-loop current sensor, the temperature sensor is a PT100 platinum resistance thermometer, the vision sensor is a Basler CMOS industrial camera equipped with a fixed-focus lens and a ring LED fill light, and the encoder is a Heidenhain incremental encoder.

Images