Intelligent evaluation system for mutual inductor

Abstract

According to the transformer intelligent evaluation system provided by the invention, information of each unit is comprehensively analyzed through the central control unit, and task instructions are dynamically generated, so that system-level coordination and decision-making are realized; the image recognition unit collects and processes visual information, the transmission mechanical arm executes precise carrying based on a processing result, the insulation test cabin completes a high-voltage insulation test, the error measurement unit achieves precise electrical parameter measurement, and the environment monitoring system monitors and adjusts internal environment parameters and feeds back data. According to the system, automatic and intelligent operation of the whole process of transformer evaluation is realized, the test efficiency and precision are remarkably improved, the adaptive capability and coordination of the system are enhanced, the operation safety and environmental stability are guaranteed, and meanwhile, the system has good expandability and maintainability.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection technology for current transformers, and particularly to an intelligent evaluation system for current transformers. Background Technology

[0002] Performance testing of high-voltage electrical equipment, especially the evaluation of insulation strength and measurement accuracy of instrument transformers, is a crucial link in ensuring the safe and stable operation of the power system. Traditional testing methods mainly rely on manual operation, where operators manually move the instrument transformers under test to the insulation test bench and error verification device station in sequence, and complete the wiring, testing, and data recording process according to regulations. This mode is not only inefficient and labor-intensive, but also prone to operational errors or judgment biases due to human factors. With the development of industrial automation and machine vision technology, semi-automated testing solutions using robotic arms for workpiece transfer have gradually emerged in the industry. However, existing technologies mostly focus on the realization of single functions, lacking system-level coordination and optimization. Information isolation between various functional units leads to low overall system efficiency, poor adaptability, and reliability that cannot meet the requirements of high-precision continuous operation. Summary of the Invention

[0003] In view of this, the present invention provides an intelligent evaluation system for current transformers to address the technical deficiencies existing in the prior art.

[0004] Specifically, the present invention provides an intelligent evaluation system for current transformers, comprising: The central control unit analyzes information from other units, dynamically generates a sequence of task instructions, and distributes them to other units to perform system-level coordination and decision-making. An image recognition unit, connected to the central control unit, is used to collect and process visual information; The robotic arm is connected to the central control unit and the image recognition unit, and is used to perform current transformer handling operations based on the processing results of the image recognition unit under the control of the central control unit. The insulation test chamber, connected to the central control unit, is used to receive current transformers placed by the transfer robotic arm and perform high-voltage insulation tests. The error measurement unit, connected to the central control unit, is used to perform precision electrical parameter measurements on the current transformer placed by the transfer robotic arm. The environmental monitoring system is connected to the central control unit to monitor and adjust the internal environmental parameters of the system and feed the environmental data back to the central control unit.

[0005] In some implementations, the central control unit integrates a task scheduling module; The task scheduling module adopts a decision model based on deep reinforcement learning. The decision model updates its internal state by continuously receiving target pose data from the image recognition unit, real-time temperature and humidity data from the environmental monitoring system, working status data from the transmission robotic arm, and working status data from the error measurement unit, and outputs the optimal task allocation strategy to the execution controller of the transmission robotic arm.

[0006] In some implementations, the transfer robotic arm is equipped with a path planning module that employs an improved search algorithm. The search algorithm obtains currently available workspace information by querying a dynamic equipment topology map maintained by a central control unit and integrates it with 3D environmental point cloud data generated in real time by an image recognition unit to construct a cost map containing dynamic obstacles, ultimately generating a collision-free and smooth motion trajectory.

[0007] In some implementations, before performing a measurement, the error measurement unit calls its built-in calibration module to execute a self-calibration procedure. The compensation coefficient used in the self-calibration procedure is obtained by the central control unit by looking up a table based on the historical temperature and humidity data reported by the environmental monitoring system, thereby eliminating system errors.

[0008] In some implementations, the system supports hot-swapping of each functional unit. When the central control unit detects that a module has been removed or connected, it triggers a device re-enumeration process, updates the dynamic device topology map, and notifies the path planning module to recalculate all affected task paths.

[0009] In some implementations, the central control unit also includes a data logging module for storing key parameters during all task execution processes. These key parameters are used for offline training and optimization of the deep reinforcement learning decision model.

[0010] In some implementations, the task scheduling module calculates the overall system performance index within each decision cycle and uses this index to guide the optimization direction of the decision model. The formula for calculating the overall system performance index includes:

[0011]

[0012] Where P is the overall system performance index; N is the total number of historical tasks recorded in the data recording module; and The pre-defined normalized weighting coefficients are used to balance the contributions of the two indicators. The ideal completion time for the k-th task, calculated based on task complexity, is obtained by the task parsing module of the central control unit. The actual completion time of the kth task recorded by the data recording module; The penalty factor, given by the system configuration parameters, is used to adjust the system's sensitivity and preference for completing tasks "timeout" or "early"; M is the total number of controllable modules in the system. and These are the maximum and minimum utilization rates of the i-th module within the statistical period, obtained from the statistics of the data recording module. A very small constant introduced to prevent division by zero errors.

[0013] In some implementations, the ideal completion time is obtained by performing critical path analysis on the task using a directed acyclic graph. The calculation formulas corresponding to the critical path analysis include:

[0014] Where Paths represents the set of all possible paths that constitute the task; p represents one of the paths; j represents a task node on path p; This represents the standard operation time estimated by the task parsing module for the j-th task node. The baseline performance coefficients of the modules that process this task node are obtained by the central control unit from the module capability description file; and The complexity adjustment factor associated with the type of the j-th task node is given by the system experience parameter table; This refers to the offset between the current environmental parameters reported by the environmental monitoring system and the standard parameters.

[0015] In some implementations, the image recognition unit integrates a depth camera and an RGB camera, and establishes a precise coordinate transformation relationship with the transmission robotic arm through a hand-eye calibration program.

[0016] In some implementations, the insulation test chamber is equipped with a safety interlock mechanism directly connected to the central control unit to ensure that high-voltage testing is initiated only when the chamber door is fully closed and the transfer robotic arm is in a safe position.

[0017] At least one embodiment of this invention achieves system-level coordination and decision-making by comprehensively analyzing information from various units and dynamically generating task instructions through a central control unit; an image recognition unit collects and processes visual information, a robotic arm performs precise handling based on the processing results, an insulation test chamber completes high-voltage insulation testing, an error measurement unit performs precise electrical parameter measurement, and an environmental monitoring system monitors and adjusts internal environmental parameters and feeds back data. This system achieves fully automated and intelligent operation of the instrument transformer evaluation process, significantly improving testing efficiency and accuracy, enhancing system adaptability and coordination, ensuring operational safety and environmental stability, and possessing good scalability and maintainability. Attached Figure Description

[0018] Figure 1 This is a structural block diagram of an intelligent evaluation system for current transformers provided by the present invention. Detailed Implementation

[0019] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0020] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the one or more embodiments of this specification. The singular forms “a” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items. The modifications “a” and “a plurality” as used in this disclosure are illustrative and not restrictive, and those skilled in the art will understand that they should be understood as “one or more” unless the context clearly indicates otherwise.

[0021] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0022] See Figure 1 , Figure 1A structural block diagram of an intelligent instrument transformer evaluation system according to some embodiments of this specification is shown. The intelligent instrument transformer evaluation system includes: a central control unit, which dynamically generates a sequence of task instructions by analyzing information from other units and distributes them to other units to perform system-level coordination and decision-making; an image recognition unit connected to the central control unit for acquiring and processing visual information; a transfer robotic arm connected to the central control unit and the image recognition unit for performing instrument transformer handling operations based on the processing results of the image recognition unit under the control of the central control unit; an insulation test chamber connected to the central control unit for receiving instrument transformers placed by the transfer robotic arm and performing high-voltage insulation tests; an error measurement unit connected to the central control unit for performing precision electrical parameter measurements on the instrument transformers placed by the transfer robotic arm; and an environmental monitoring system connected to the central control unit for monitoring and adjusting internal environmental parameters of the system and feeding back environmental data to the central control unit.

[0023] The central control unit can refer to the hardware and software modules that serve as the core processing and coordination center of the system. For example, this unit may employ a multi-core industrial-grade processor and exchange data and distribute instructions with other units via Ethernet and real-time bus protocols. It is used to coordinate the workflow of the entire system and make intelligent decisions. Dynamically generating task instruction sequences refers to the process of automatically generating an ordered set of operation commands based on the real-time system status. For example, based on image recognition results and environmental data, a rule-based inference engine can generate instruction sequences in real time for tasks such as robotic arm movement and test chamber activation, enabling flexible task execution. System-level coordination and decision-making refers to the function of resource allocation and operational strategy formulation at the overall system level. For example, a central scheduling algorithm can synchronize the operation timing and state switching of the robotic arm, test chamber, and measurement unit to ensure the efficient and conflict-free operation of the entire evaluation process.

[0024] An image recognition unit can refer to a combination of hardware and software specifically designed for acquiring and processing visual information. For example, this unit might include a high-resolution CCD camera and an image processing algorithm based on a convolutional neural network (CNN) to accurately identify the position and orientation of the transformer. Visual information can refer to image data about the shape, position, and features of an object acquired through optical devices, such as multi-angle grayscale images and depth point cloud data of the transformer acquired by a camera, to provide the robotic arm with grasping coordinates and orientation references.

[0025] A transfer robotic arm can refer to a multi-degree-of-freedom robotic device capable of automatically transporting current transformers. For example, it may employ a six-axis articulated structure, driven by servo motors and reducers, and equipped with force / torque sensors, for precisely transferring current transformers between multiple workstations. Current transformer handling operations can refer to the process of moving and placing current transformers from one location to another. For instance, a robotic arm may use a vacuum suction cup or custom-designed gripper to grasp a current transformer and move it along a planned path to a predetermined position within an insulation test chamber, ready for subsequent testing.

[0026] An insulation test chamber can refer to a sealed safety container used for high-voltage insulation testing. For example, the chamber consists of a metal shielding layer, an insulating support, and high-voltage electrodes, and can apply power frequency or impulse high voltage to test the insulation strength and dielectric loss of the instrument transformer.

[0027] High-voltage insulation testing refers to the test of applying high voltage to electrical equipment to check its insulation performance. For example, according to the IEC 60060 standard, AC high voltage is applied between the primary winding of the instrument transformer and ground and the leakage current is monitored to determine whether the insulation is qualified.

[0028] An error measurement unit can refer to equipment used for precise measurement of the electrical parameters of a current transformer. For example, this unit may include a standard voltage source, a proportional standard, and a phase comparator, using the difference method to measure the ratio difference and phase difference to assess the accuracy class of the current transformer. Precision electrical parameter measurement can refer to the high-precision quantification process of parameters such as amplitude, phase, and frequency of electrical signals. For example, under rated load, multiple standard voltages and currents are applied, and input and output signals are simultaneously acquired and errors are calculated to obtain the ratio difference and phase difference data of the current transformer.

[0029] An environmental monitoring system can refer to a combination of devices that continuously monitor and adjust the internal environmental conditions of a system. For example, it might use temperature and humidity sensors, data acquisition units, and PLC controllers to monitor and adjust the temperature and humidity of the test area in real time to ensure the accuracy and repeatability of test results. Internal environmental parameters can refer to the physical condition data of the space in which the system operates, such as ambient temperature, relative humidity, atmospheric pressure, and cleanliness level. This data is collected through a sensor network and uploaded to the central controller. Environmental data can refer to a set of measurements describing the physical conditions around the system. For example, analog temperature and humidity signals collected in real time by a PT100 temperature sensor and a capacitive humidity sensor, converted into digital data by an analog-to-digital converter, are used to provide environmental status input for system decision-making.

[0030] The present invention will be further described below through a detailed embodiment: The intelligent transformer evaluation system of the present invention is deployed on the fully automated verification line of a power metering and testing center for unmanned performance evaluation of 0.2S class voltage transformers.

[0031] The core of the system is the central control unit, which adopts a multi-core processor architecture and establishes physical connections with each subsystem through a gigabit industrial Ethernet switch. It runs a scheduling system based on a Linux real-time kernel. By parsing the work orders issued by the MES (Manufacturing Execution System), it dynamically generates a sequence of task instructions, including the robotic arm path point sequence, test chamber pressurization parameters, and measurement unit range instructions, and distributes them to each execution unit using the Modbus-TCP industrial communication protocol.

[0032] The image recognition unit is equipped with a 2448×2048 resolution global shutter CMOS sensor and an 850nm wavelength active infrared light source, continuously acquiring visual data from the current transformers on the transmission line via the GigE Vision protocol. Its built-in deep learning model adopts the YOLOv5 architecture and can output the three-dimensional coordinates and deflection angle of the center point of the current transformer flange in real time, with a positioning accuracy of ±0.15 mm.

[0033] The robotic arm is a six-axis articulated structure with a repeatability of ±0.05 mm. Its end effector is a non-contact vacuum suction cup based on Bernoulli's principle. It receives target pose commands from the central control unit via an EtherCAT bus and simultaneously acquires real-time coordinate data output from the image recognition unit. Through inverse kinematics calculations, it generates servo control signals for each joint angle, completing the grasping and transporting operation of the current transformer.

[0034] The insulation test chamber employs a power frequency withstand voltage test system with a rated voltage of 80 kV, and the inner wall of the chamber is covered with a 2 mm thick equalizing shielding layer. After the robotic arm places the current transformer on the insulation platform with a rated load of 50 kg, the pneumatic locking device of the chamber door is activated. After the central control unit detects the dual ready signals from the door limit switch and the robotic arm safety position sensor through the digital I / O module, it triggers the test program to automatically apply a voltage increase rate of 42 kV / min until the rated test voltage is reached.

[0035] The error measurement unit integrates a 0.005-level standard source and a 24-bit high-precision ADC (analog-to-digital converter) acquisition card, covering a measurement range of 50V to 690V power frequency voltage. It receives test point commands from the central control unit via the GPIB bus, applying 20%, 100%, and 120% of the rated voltage sequentially under a 0.8 power factor condition, simultaneously acquiring the secondary voltage difference between the standard transformer and the transformer under test, and calculating the ratio difference and phase difference.

[0036] The environmental monitoring system is equipped with a PT1000 platinum resistance temperature sensor and a capacitive humidity sensor, covering a monitoring range of -10℃ to +50℃ and 5%RH to 95%RH. It uploads real-time temperature and humidity data to the central control unit via a 4-20mA analog circuit. When environmental parameters exceed the rated operating range of 23±2℃ and 55±10%RH, the environmental control system, consisting of the industrial air conditioning unit and humidifier, is triggered to activate its compensation mechanism.

[0037] During system operation, the central control unit continuously monitors the status registers of each subsystem. Once the image recognition unit confirms through visual processing that the current transformer has entered the working field of view, the system automatically triggers the robotic arm grasping process. After the error measurement unit completes the data acquisition of all test points, it automatically generates a test report conforming to the DL / T 668 standard and uploads it to the quality management system. The entire process requires no manual intervention, and the evaluation cycle for a single current transformer is controlled within 8 minutes, improving efficiency by more than 300% compared to traditional manual operation.

[0038] The beneficial effects of one of the embodiments in this specification include at least the following: The central control unit comprehensively analyzes information from each unit and dynamically generates task instructions, achieving system-level coordination and decision-making; the image recognition unit collects and processes visual information, and the robotic arm performs precise handling based on the processing results; the insulation test chamber completes high-voltage insulation testing; the error measurement unit achieves precise electrical parameter measurement; and the environmental monitoring system monitors and adjusts internal environmental parameters and feeds back data. This system achieves fully automated and intelligent operation of the instrument transformer evaluation process, significantly improving testing efficiency and accuracy, enhancing system adaptability and coordination, ensuring operational safety and environmental stability, while also possessing good scalability and maintainability.

[0039] In some implementations, the central control unit integrates a task scheduling module; the task scheduling module adopts a decision model based on deep reinforcement learning. The decision model updates its internal state by continuously receiving target pose data from the image recognition unit, real-time temperature and humidity data from the environmental monitoring system, working status data from the transmission robotic arm, and working status data from the error measurement unit, and outputs the optimal task allocation strategy to the execution controller of the transmission robotic arm.

[0040] The task scheduling module can refer to the functional unit in the central control unit responsible for resource allocation and task sequencing. For example, this module might employ a reinforcement learning algorithm based on Deep Q-Network (DQN), with the input state space including robotic arm joint angles and visual recognition confidence levels, to dynamically optimize the task execution order of each unit to reduce idle waiting time. A deep reinforcement learning-based decision model can refer to an adaptive decision framework that combines deep neural networks with reinforcement learning. For example, this model takes the state vectors of each unit as input and outputs an action value function through a policy network to autonomously learn the optimal task scheduling strategy. Target pose data can refer to a set of parameters describing the target's position and orientation in three-dimensional space, such as homogeneous transformation matrices containing X, Y, and Z coordinate values ​​and rotation angles around each axis, calculated by the image recognition unit through hand-eye calibration. Real-time temperature and humidity data can refer to the ambient temperature and humidity measurements collected at the current moment, such as digital data read from an SHT35 sensor via an I2C bus, with a sampling frequency of 1Hz and an accuracy of ±0.2℃ for temperature and ±1.5%RH for humidity. Operating status data can refer to a set of parameters reflecting the current operating status of the equipment, such as structured data frames containing servo drive enable status, motor current values, and fault codes, which are periodically transmitted via the EtherCAT bus. Internal state can refer to intermediate variables and memory information maintained by the decision model during operation, such as the activation values ​​of hidden layer neurons and the recurrent network state vector in a DRL model, used to record historical state change trajectories. Optimal task allocation strategy can refer to a resource allocation scheme that maximizes the overall system performance, such as prioritizing insulation testing tasks for the output robotic arm and enabling a fast calibration mode for the error measurement unit to minimize the total task completion time. The execution controller can refer to the underlying control unit that drives the servo motors in the robotic arm, such as a servo drive using the CIA 402 protocol standard, which receives position mode commands and performs closed-loop control of the motor's rotation angle.

[0041] As a concrete example: On a testing production line, the task scheduling module starts working after the system boots up. The module's built-in deep reinforcement learning decision model (using a dual-objective network architecture) continuously receives input at a frequency of 10Hz: target pose data (including the 6-dimensional pose vector of the current transformer's grasping point) sent by the image recognition unit via a gigabit Ethernet port; real-time temperature and humidity data (temperature 23.5℃, humidity 48%RH) sent by the environmental monitoring system via an RS485 bus; working status data (including the enable status of each joint and actual position values) sent by the robotic arm via an EtherCAT fieldbus; and self-test status codes sent by the error measurement unit via PROFINET. The decision model updates its internal state (including 128-dimensional LSTM hidden states), and after calculation by the policy network, outputs the optimal task allocation strategy: prioritizing the insulation test of the current transformer at station 3, while simultaneously scheduling the robotic arm to grasp the product to be inspected at station 5. This strategy is sent to the execution controller of the robotic arm via the Modbus-TCP protocol. After parsing the instructions, the controller drives the servo motor to complete the corresponding actions.

[0042] The intelligent task scheduling mechanism optimizes the allocation of system resources, significantly improves equipment utilization and overall operational efficiency, ensures the efficient and coordinated execution of complex task sequences, and enhances the system's adaptability to dynamic environments.

[0043] In some implementations, the transfer robotic arm is equipped with a path planning module that employs an improved search algorithm. The search algorithm obtains currently available workspace information by querying a dynamic equipment topology map maintained by a central control unit and integrates it with 3D environmental point cloud data generated in real time by an image recognition unit to construct a cost map containing dynamic obstacles, ultimately generating a collision-free and smooth motion trajectory.

[0044] The path planning module can refer to the algorithm unit in the robotic arm responsible for generating motion trajectories; for example, this module may employ an improved RRT. The Rapidly-exploring Random Tree Star (RRT) algorithm takes the robotic arm's DH parameters and obstacle point clouds as input and calculates a collision-free and energy-optimal motion path. Improved search algorithms can refer to random sampling algorithms optimized from traditional path search methods, such as those used in RRT. The algorithm incorporates dynamic step size adjustment and heuristic biased sampling strategies to improve search efficiency and path quality in complex environments. The dynamic device topology map refers to a spatial structure diagram that reflects the real-time positional relationships of various system units. For example, it uses a directed graph data structure to store device node coordinates and connection states, and maintains data consistency across multiple terminals through a version number mechanism. Currently available workspace information refers to the spatial range parameters reachable by the robotic arm's end effector, such as a three-dimensional spatial description matrix including joint movement limits, singular point locations, and interference regions, used to constrain the path search domain. Real-time generated 3D environmental point cloud data refers to instantaneous three-dimensional spatial information acquired through depth sensors. For example, it uses a structured light camera to output 1280×720 resolution point cloud data at a frequency of 30Hz, with each point containing XYZ coordinates and reflection intensity information. The cost map containing dynamic obstacles refers to a rasterized environmental map incorporating information about moving obstacles. For example, it overlays the predicted trajectory of a moving person onto a static map as a time-dilated layer to assess the collision risk of path points. Collision-free and smooth motion trajectories can refer to robotic arm paths that avoid obstacles and are continuously differentiable. For example, B-spline curves can be used to smooth the original path to ensure that the acceleration of each joint is continuous and without abrupt changes.

[0045] As a concrete example: when the robotic arm receives a handling instruction, its path planning module starts running. This module calls an improved RRT (Real-Time Tracking) function. The algorithm first queries the dynamic equipment topology map maintained by the central control unit via the OPC UA protocol to obtain the status of the insulation test chamber door (opening angle 70 degrees) and the location information of the error measurement unit within the current work area. Simultaneously, it receives real-time 3D environmental point cloud data (including the point cloud of a mobile tool vehicle that suddenly enters the safe zone) transmitted by the image recognition unit via gigabit Ethernet. The algorithm fuses the static equipment coordinates and dynamic obstacle information to generate a cost map, where the cost value for the mobile tool vehicle area is set to 255, and for the static equipment area, it is set to 100. After 500 iterations of sampling, a collision-free trajectory containing 7 path points is generated. Finally, cubic B-spline interpolation is used to ensure trajectory smoothness, and the maximum acceleration of each joint does not exceed a set threshold.

[0046] Intelligent path planning enables the robotic arm to move safely and efficiently in dynamic environments, effectively avoiding the risk of collisions with static equipment and dynamic obstacles, ensuring the smoothness and accuracy of the handling process, and improving the system's adaptability to environmental changes.

[0047] In some implementations, before performing a measurement, the error measurement unit calls its built-in calibration module to execute a self-calibration procedure. The compensation coefficient used in the self-calibration procedure is obtained by the central control unit by looking up a table based on the historical temperature and humidity data reported by the environmental monitoring system, thereby eliminating system errors.

[0048] Calling its built-in calibration module can refer to activating the standard comparison function integrated within the error measurement unit. For example, it can initiate the output of a standard signal from the built-in reference source by sending a specific SCPI (Standard Commands for Programmable Instruments) command for self-calibration. The self-calibration procedure refers to the standardized process by which the device automatically completes accuracy verification. For example, it can perform zero-point calibration, range calibration, and linearity verification sequentially according to the steps specified in the IEC 60500 standard to ensure that the measurement accuracy meets the requirements. The compensation coefficient refers to the adjustment parameters used to correct system measurement errors. For example, it can be a digital matrix containing gain correction factors and offset compensation values, calculated using a polynomial fitting algorithm, used to correct errors in the original measurement data. The reported historical temperature and humidity data refers to past environmental records periodically collected and uploaded by the environmental monitoring system. For example, it can be a data structure containing timestamps, temperature sampling values, and humidity sampling values, uploaded to the central database at 1-minute intervals via the Modbus protocol. Eliminating system errors refers to reducing or removing the inherent bias components of the measuring equipment. For example, it can apply compensation coefficients to mathematically transform the original sampled values, making the measurement results closer to the true value and improving measurement accuracy.

[0049] As a concrete example: when the error measurement unit prepares to perform a ratio difference measurement on a 0.2-class voltage transformer, it first invokes its built-in calibration module. This calibration module automatically executes a self-calibration procedure: controlling the internal 0.05-class standard source to output a 100V / 50Hz reference voltage, acquiring the actual output value through a 24-bit high-precision ADC, and calculating the gain and offset error. Simultaneously, the central control unit, based on historical temperature and humidity data reported by the environmental monitoring system (temperature variation of 22.5-23.8℃ and humidity of 45-52%RH within the last hour), retrieves the corresponding temperature compensation coefficient α=0.00015 and humidity compensation coefficient β=0.00008 from a pre-stored compensation coefficient table. These coefficients are downloaded to the measurement unit's DSP processor, correcting the measured values ​​in real time during the calibration process, ultimately controlling the system error within ±0.005%.

[0050] The adaptive calibration mechanism significantly improves measurement accuracy and reliability, effectively eliminates systematic errors caused by environmental factors, ensures the consistency and accuracy of measurement results, and reduces reliance on manual intervention and external standards.

[0051] In some implementations, the system supports hot-swapping of each functional unit. When the central control unit detects that a module has been removed or connected, it triggers a device re-enumeration process, updates the dynamic device topology map, and notifies the path planning module to recalculate all affected task paths.

[0052] Hot-swapping capability refers to the ability to connect and disconnect modules while they are powered on during system operation. This can be achieved through methods such as using a PICMG-compliant backplane bus design and hot-swappable controller chips to support plug-and-play and hot-swap functionality. Detecting module removal or connection refers to the central control unit's awareness of changes in system hardware configuration. This can be achieved by monitoring changes in PCIe link training status and power management signals to detect changes in module connection status in real time. Triggering device re-enumeration refers to initiating a standardized process for re-identifying and registering hardware devices. This can be achieved by calling the Linux kernel's udev device management mechanism to rescan device nodes on the bus and update the device tree information. Updating the dynamic device topology graph refers to refreshing the device relationship data structure maintained internally by the system. This can be achieved by using a graph database to store device connection relationships and automatically updating the adjacency matrix when a device change is detected. Recalculating all affected task paths refers to generating new motion trajectories based on the new system configuration. For example, when the end effector of a robotic arm is replaced, kinematic solutions and trajectory planning are re-performed based on the new DH parameters and workspace constraints.

[0053] As a concrete example: During system operation, when technicians need to replace the data acquisition card module of the error measurement unit, they directly perform a hot-swap operation. The central control unit detects the removal of a module by monitoring the PRSNT# pin level change on the PCIe bus. The system immediately triggers the device re-enumeration process: it calls the udevadm tool on the Linux system to rescan the PCI bus devices and identifies the newly inserted data acquisition card hardware ID as 0x1234. Subsequently, it updates the dynamic device topology map, updating the device node information, address range, and performance parameters of the new data acquisition card to the system configuration database. The central control unit sends a device change notification to the path planning module through the ROS / pub topic publishing mechanism. After receiving the notification, the path planning module recalculates all task paths involving the measurement unit based on the new device layout information, including the target pose and obstacle avoidance trajectory for the robotic arm to place the current transformer, ensuring the continued safe and stable operation of the system.

[0054] The hot-swappable function enables high availability and easy maintenance of the system, allowing modules to be replaced and upgraded without shutting down the system. This significantly reduces system maintenance time, improves equipment utilization and operational flexibility, and ensures the continuous and stable operation of the system after configuration changes.

[0055] In some implementations, the central control unit also includes a data logging module for storing key parameters during all task execution processes. These key parameters are used for offline training and optimization of the deep reinforcement learning decision model.

[0056] A data logging module can refer to a hardware or software component specifically responsible for storing operational data. For example, it might use an SSD as the storage medium and directly record the status data and measurement results of each unit via DMA to save historical information about the system's operation. Key parameters can refer to monitoring data that significantly impacts system performance and task completion quality. These could be structured datasets containing task start timestamps, unit status codes, environmental monitoring values, and final measurement results, used to evaluate the system's operational status. A deep reinforcement learning decision model can refer to a self-learning model that combines deep neural networks with reinforcement learning algorithms. For example, it could use the DQN algorithm architecture, taking the system state vector as input and outputting the optimal action policy, to achieve intelligent task scheduling.

[0057] As a concrete example: During daily system operation, the data logging module works continuously. This module connects to an NVMe SSD storage device via a PCIe interface, recording key parameters of each unit at 10ms intervals: acquiring joint angle and torque data from the robotic arm controller, recording pressurization time and leakage current values ​​from the insulation test chamber, storing raw readings of specific difference and angle difference from the error measurement unit, and simultaneously recording temperature and humidity data from the environmental monitoring system. All data is timestamped and stored in a custom binary format. During weekly system maintenance, engineers export this historical data and use the TensorFlow framework to train a deep reinforcement learning decision model offline: using historical state data as input and actual task completion time as a reward signal, the network parameters are optimized through experience replay and gradient descent algorithms, enabling the model to learn more efficient task scheduling strategies.

[0058] By establishing a comprehensive data recording and model optimization mechanism, the system's performance has been continuously improved and it has achieved self-evolution, providing reliable data support for intelligent decision-making and enhancing the system's learning adaptability and long-term operational stability.

[0059] In some implementations, the task scheduling module calculates the overall system performance index within each decision cycle and uses this index to guide the optimization direction of the decision model. The formula for calculating the overall system performance index includes:

[0060]

[0061] Where P is the overall system performance index; N is the total number of historical tasks recorded in the data recording module; and The pre-defined normalized weighting coefficients are used to balance the contributions of the two indicators. The ideal completion time for the k-th task, calculated based on task complexity, is obtained by the task parsing module of the central control unit. The actual completion time of the kth task recorded by the data recording module; The penalty factor, given by the system configuration parameters, is used to adjust the system's sensitivity and preference for completing tasks "timeout" or "early"; M is the total number of controllable modules in the system. and These are the maximum and minimum utilization rates of the i-th module within the statistical period, obtained from the statistics of the data recording module. A very small constant introduced to prevent division by zero errors.

[0062] The decision cycle can refer to the fixed or variable time interval between two consecutive decisions made by the system. For example, the central control unit calls the inference function of the task scheduling module at a fixed frequency (e.g., 10 times per second), inputs the latest system state, and outputs decision instructions, ensuring that the system can respond promptly to changes in internal state. The comprehensive system performance index can refer to a composite value that quantitatively evaluates the overall operating efficiency and resource utilization of the system. For example, it can be composed of a weighted sum of squares operation of task efficiency sub-items and equipment utilization balance sub-items, comprehensively measuring system performance and providing clear targets for optimization. The total number of historical tasks can refer to the number of completed tasks stored in the system data recording module. For example, it can be used as a normalization factor N in the performance index calculation formula to calculate the average level of task efficiency, eliminating the impact of fluctuations in the number of tasks on the evaluation results. The normalization weighting coefficient can refer to a scaling factor used in multi-objective optimization formulas to balance the contribution of different sub-items. For example, it can be preset by the system administrator according to business needs. and The value (e.g.) =0.7, =0.3), to adjust the relative importance of task completion efficiency and equipment utilization in the comprehensive index. Ideal completion time can refer to the theoretical shortest estimated time required to complete a specific task under ideal conditions. For example, it can be calculated by performing critical path analysis on the directed acyclic graph corresponding to the task and combining standard operation time, module baseline performance and environmental offset, and is used as a benchmark for evaluating the actual task execution efficiency.

[0063] Actual completion time refers to the actual time spent from the start of task execution to its final completion. For example, it can be calculated by the data recording module adding a timestamp when the task status changes to "complete" and then subtracting the task start timestamp. This difference is used to compare the time with the ideal completion time to calculate task execution efficiency. A weighting index can refer to an exponential constant used to adjust the relative importance of a certain term in the calculation formula. For example, when calculating the task efficiency sub-item, the index... Used to amplify or reduce the difference between actual and ideal time, allowing the system to focus more on optimizing tasks that exceed the ideal time significantly. The total number of controllable modules can refer to the number of functional units in the system whose operating status can be monitored and adjusted by the central control unit. For example, it can be used as a normalization factor M when calculating the equipment utilization balance sub-item, representing the size of the set of modules whose utilization needs to be monitored. Maximum utilization can refer to the peak value of a module's resource utilization within a specific statistical period. For example, a data logging module continuously samples CPU utilization, memory usage, etc., and records their maximum values. This is used to characterize the highest load level of the module within a period. Minimum utilization can refer to the lowest value of resource utilization of a module within a specific statistical period. For example, a data logging module continuously samples the resource indicators of each module and records its minimum value. This is used to characterize the idle or low-load state of the module within a period. The statistical period can refer to a time window used to aggregate and calculate statistical indicators such as utilization. For example, the data logging module maintains time-series data on resource utilization for each module in a rolling window manner (e.g., the most recent 24 hours), providing a data foundation for calculating maximum and minimum utilization. Preventing division by zero errors can refer to protective measures taken in mathematical calculations to avoid calculation errors caused by a denominator of zero. For example, adding a very small constant ε (e.g., 1e-10) to the denominator of the calculation formula ensures that the formula can still be calculated effectively when U_i^min is zero.

[0064] As a concrete example: When a decision cycle is triggered, the DRL (Deep Reinforcement Learning) agent of the task scheduling module is invoked. Its state input s_t is a multi-dimensional vector composed of concatenations, including the latest target pose data sent by the image recognition unit (a set of 6-dimensional SE(3) poses), the real-time temperature and humidity data reported by the environmental monitoring system (temperature 25.1℃, humidity 45%RH), the joint angle and status flag of the transmission robot arm, and the current working status of the error measurement unit (idle / busy / calibrating). The agent outputs action a_t based on its policy network π(a|s), which is a multi-dimensional continuous action vector. This vector is decoded by the execution controller into a specific task allocation instruction for the next period, such as instructing the transmission robot arm to move the current transformer that has completed the insulation test to the error measurement unit. At the same time, the data recording module records the system state, decision actions, and subsequent task completion times and other key parameters in the time series database during this cycle. At the end of the cycle, the system's overall performance index P will be calculated according to the latest recorded data and a given formula. The result of this calculation will be used as part of the reward signal r_t for subsequent offline training and parameter updates of the DRL decision model.

[0065] By incorporating both task execution efficiency and the balance of equipment resource utilization into a quantifiable comprehensive performance index, and designing a refined calculation formula for this index, the task scheduling module based on deep reinforcement learning can obtain a comprehensive, stable, and clearly oriented reward signal. This guides the agent to learn an efficient scheduling strategy that not only completes tasks quickly but also maintains a balanced load across all modules of the system, ultimately improving the throughput and stability of the entire evaluation system.

[0066] In some implementations, the ideal completion time is obtained by performing critical path analysis on the task using a directed acyclic graph. The calculation formulas corresponding to the critical path analysis include:

[0067] Where Paths represents the set of all possible paths that constitute the task; p represents one of the paths; j represents a task node on path p; This represents the standard operation time estimated by the task parsing module for the j-th task node. The baseline performance coefficients of the modules that process this task node are obtained by the central control unit from the module capability description file; and The complexity adjustment factor associated with the type of the j-th task node is given by the system experience parameter table; This refers to the offset between the current environmental parameters reported by the environmental monitoring system and the standard parameters.

[0068] A directed acyclic graph (DAG) can refer to a graph data structure consisting of nodes and directed edges without any cycles. For example, a task parsing module decomposes a high-level task into a series of atomic operation nodes with dependencies, and connects these nodes with directed edges to represent the execution order, clearly modeling the internal logic and execution flow of complex tasks. Critical path analysis (CPA) is a technique used in project management to determine the shortest completion time of a project and the sequence of tasks that directly affect the overall project duration. For example, by topologically sorting the task DAG and using dynamic programming to calculate the total time of all paths, the path with the maximum value is selected to identify the core operation chain that determines the ideal completion time of the task. A task node can refer to the basic unit that constitutes the task DAG, representing an indivisible atomic operation or processing step, such as "robotic arm grasping the current transformer," "pressurizing the insulation test chamber," and "error measurement unit performing measurement," and is the basic element for calculating the total time of a path. Standard operation time refers to the expected time to execute a specific task node under standard experimental conditions. For example, the task parsing module pre-assigns a time value ω_j to each type of task node based on historical statistical data or the technical specifications provided by the module manufacturer, serving as the basic input parameter for calculating the ideal completion time. Baseline performance coefficient refers to a dimensionless scaling factor characterizing the relative performance level of a functional module when processing its specific type of task. For example, the central control unit queries the nominal processing speed ν_j of each module from its XML description file, which is used to scale the standard operation time in the denominator when calculating the ideal completion time. Module capability description file refers to a configuration file stored in a structured data format, such as XML or JSON, that describes the various performance indicators and functions of a functional module. It contains information such as the module ID, a list of supported task types, the baseline performance coefficient ν_j, and interface protocols, which can be queried and parsed by the central control unit. Complexity adjustment factors refer to two coefficients, such as β_j and γ_j, used to fine-tune the time estimation of a task node based on its specific attributes when calculating the ideal completion time. Their specific values ​​are obtained from an empirical parameter table maintained by the system based on task type and historical experience, so as to more accurately reflect the execution difficulty of different nodes.

[0069] Task node type refers to the classification of task nodes according to their operational nature or required resources, such as "vision processing," "mechanical motion," "high-precision measurement," and "high-pressure testing." Different types of nodes correspond to different complexity adjustment factors and empirical parameters. The system empirical parameter table refers to a data table maintained by the system that stores empirical adjustment parameters for various task nodes under different conditions. For example, it could be a lookup table stored in a database, where the key is the task node type and the value is the corresponding... and The parameter set provides empirical data support for calculating ideal time. Environmental parameters can refer to physical quantities describing the internal working environment of the system, such as temperature, humidity, air pressure, and cleanliness. These parameters are continuously monitored and reported by various sensors (such as temperature and humidity sensors) in the environmental monitoring system. Standard parameters can refer to pre-set reference values ​​or ideal ranges of environmental parameters to ensure the normal operation of each functional module of the system, such as a standard temperature of 20 degrees Celsius and a standard relative humidity of 50%, serving as a benchmark for assessing whether the current environment deviates from the ideal state. Offset can refer to the difference between the current measured value of a certain environmental parameter and the standard parameter value set for it. For example, the environmental monitoring system calculates the difference between the current temperature of 25.1℃ and the standard temperature of 20℃ and reports such offsets to the central control unit for calculating the impact of the environment on task time.

[0070] As a concrete example: When the task parsing module of the central control unit receives a high-level instruction to "complete a comprehensive test of a certain type of current transformer," it first parses it into a specific directed acyclic graph (DAG). This DAG contains multiple task nodes, such as "positioning and grabbing," "transferring to the insulation chamber," "performing insulation testing," "transferring to the error measurement platform," and "performing ratio / angle difference measurement," etc. The directed edges between nodes define strict execution order dependencies (e.g., the error measurement can only be performed after the insulation test is completed). Subsequently, the module performs critical path analysis on each path from start to finish. For each task node j on path p, the module queries its capability description file (XML format) for the baseline performance coefficients of the modules handling tasks of that node type. It also reads the complexity adjustment factor corresponding to the node type from the system experience parameter database table. and Meanwhile, the environmental monitoring system reports the deviations of the current ambient temperature and humidity from the standard values ​​via the Modbus protocol. The time taken for this path is also obtained. Finally, the ideal time for this path is calculated using the aforementioned formula, and the maximum time taken for all paths is determined as the ideal completion time for the task. The data is then stored in the data recording module for use by the performance indicator calculation module.

[0071] By introducing critical path analysis based on directed acyclic graphs to calculate the ideal completion time of tasks, and comprehensively considering the theoretical performance of modules, the inherent complexity of task nodes, and the quantitative impact of real-time environmental conditions on execution efficiency, the resulting ideal time estimates are more accurate and in line with theoretical optimality. This provides a scientific and reliable benchmark for calculating the overall system performance index, making the scheduling optimization direction based on this index more accurate and effective.

[0072] In some implementations, the image recognition unit integrates a depth camera and an RGB camera, and establishes a precise coordinate transformation relationship with the transmission robotic arm through a hand-eye calibration program.

[0073] A depth camera can refer to a visual sensor capable of acquiring distance information for each pixel in a scene. For example, it might use structured light or time-of-flight principles to output point cloud data containing X, Y, and Z coordinates, providing 3D geometric information of the target object. An RGB camera can refer to a color visual sensor capable of capturing the red, green, and blue primary colors of a scene. For example, it might output a standard 2D color image to provide appearance information such as texture and color of the target. A hand-eye calibration procedure can refer to a set of mathematical calculations and operational processes that determine the coordinate transformation relationship between the visual sensor and the robot actuator. For example, it might involve moving a calibration plate with a specific pattern at the end of a robotic arm across multiple positions, simultaneously acquiring images and joint angles, and finally solving for the transformation matrix between the camera coordinate system and the robotic arm's base coordinate system. A coordinate transformation relationship can refer to the mathematical relationship between point coordinates mapped between different coordinate systems. For example, it can be described by a 4x4 homogeneous transformation matrix, which includes rotation and translation parameters, transforming the coordinates of a point in the depth camera coordinate system to the robotic arm's base coordinate system.

[0074] As a concrete example: After system initialization or camera position change, a hand-eye calibration procedure needs to be executed to establish a precise coordinate transformation relationship between the image recognition unit and the robotic arm. The operator initiates the calibration process through the HMI interface of the central control unit. The robotic arm moves the checkerboard calibration plate fixed on its end flange to multiple (e.g., 20) different poses within the camera's field of view, according to a pre-programmed path. At each pose point, the RGB camera captures a color image of the calibration plate, and the corner pixel coordinates are extracted using the `findChessboardCorners` function in the OpenCV library; simultaneously, the depth camera captures the corresponding point cloud data; the robotic arm controller reports its current joint angle values ​​to the central control unit via the EtherCAT protocol, and the central control unit calculates the pose of the end-eye calibration plate in the robotic arm's base coordinate system using forward kinematics. Finally, the central control unit calls a hand-eye calibration solution algorithm (e.g., AX=XB) to solve for the precise homogeneous transformation matrix from the camera coordinate system to the robotic arm's base coordinate system using the collected data. This matrix is ​​persistently stored and used for all subsequent vision-guided handling operations.

[0075] By integrating depth and RGB cameras and performing precise hand-eye calibration, a high-precision spatial mapping relationship was established between the visual perception system and the robotic arm execution system. This ensures that vision-based pose measurement results can be accurately converted into the robotic arm's motion commands, thus providing a fundamental technical guarantee for the accurate and reliable grasping and placement of the mutual inductor.

[0076] In some implementations, the insulation test chamber is equipped with a safety interlock mechanism directly connected to the central control unit to ensure that high-voltage testing is initiated only when the chamber door is fully closed and the transfer robotic arm is in a safe position.

[0077] A safety interlock mechanism can refer to a set of protective devices that are electrically and / or mechanically linked to enforce the operating sequence and eliminate potential safety hazards. For example, it may be a hard-wired series circuit including door limit switches, robotic arm position sensors, and a high-voltage generator enable circuit, ensuring that the start-up conditions for high-voltage testing are absolutely met. "Completely closed door" means that the access door of the insulation test chamber has reached the designed airtight state. This can be achieved by detecting its mechanical position through a high-reliability limit switch installed on the door frame and sending a "door closed" status signal to the safety interlock logic circuit via dry contacts. This is one of the necessary prerequisites for starting the high-voltage test. A safe position can refer to a preset physical position of the transmission robotic arm, where its body and end effector are outside the influence range of the high-voltage electric field of the insulation test chamber. For example, the robotic arm controller can confirm that the center point coordinates of its end effector are within a safe zone cube by reading the encoder values ​​of each joint and performing forward kinematic calculations. This prevents the robotic arm from being subjected to electric shock or interference during high-voltage testing. High-voltage testing can refer to the process of applying a test voltage much higher than the rated operating voltage to electrical equipment to assess its insulation performance. For example, the control system in the insulation test chamber gradually increases the voltage and maintains it for a period of time according to the preset test procedure (such as IEC 60060-1 standard), while monitoring the leakage current to verify whether the insulation of the transformer is intact.

[0078] As a concrete example: After the robotic arm places a current transformer under test into the insulation test chamber and completely withdraws, the operator initiates a "Start Insulation Test" command on the HMI. The central control unit first queries the status word of the safety interlock mechanism. This status word is generated by a hardwired circuit; bit 0 is determined by the state of the normally closed contact of the door limit switch (1 represents the door is closed), and bit 1 is determined by the current status word reported by the robotic arm controller via the PROFINET protocol (1 represents being in a predefined safe position). Only when the status word is all 1s (i.e., 0x0003) does the central control unit send a high-voltage start command to the PLC of the insulation test chamber via the Modbus TCP protocol. After receiving the command, the PLC first checks the status of the local door switch and the robotic arm position safety relay again. After confirming that everything is correct, it closes the main circuit contactor and begins the pressure test according to the preset test voltage curve (e.g., 0 kV -> 3 kV / s -> 80 kV -> hold for 60s). If any bit of the safety interlock status word becomes 0 during this period, the high-voltage circuit will be immediately cut off.

[0079] By establishing a hard-wired safety interlock mechanism directly connected to the central control unit, and by forcibly associating the high-voltage test activation logic with the hatch status and robotic arm position, a safety protection layer independent of the upper-level software control logic is constructed. This fundamentally eliminates the possibility of accidentally activating the high-voltage test when personnel or equipment are in dangerous positions due to software failure or misoperation, greatly improving the safety of the entire system operation.

[0080] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this invention. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent evaluation system for current transformers, characterized in that, include: The central control unit analyzes information from other units, dynamically generates a sequence of task instructions, and distributes them to other units to perform system-level coordination and decision-making. An image recognition unit, connected to the central control unit, is used to collect and process visual information; A transfer robotic arm is connected to the central control unit and the image recognition unit, and is used to perform current transformer handling operations based on the processing results of the image recognition unit under the control of the central control unit; An insulation test chamber, connected to the central control unit, is used to receive current transformers placed by the transmission robotic arm and perform high-voltage insulation tests. An error measurement unit, connected to the central control unit, is used to perform precision electrical parameter measurements on the current transformer placed by the transmission robotic arm. An environmental monitoring system, connected to the central control unit, is used to monitor and adjust the internal environmental parameters of the system and feed the environmental data back to the central control unit.

2. The system according to claim 1, characterized in that, The central control unit integrates a task scheduling module; The task scheduling module adopts a decision model based on deep reinforcement learning. The decision model updates its internal state by continuously receiving target pose data from the image recognition unit, real-time temperature and humidity data from the environmental monitoring system, working status data from the transmission robotic arm, and working status data from the error measurement unit, and outputs the optimal task allocation strategy to the execution controller of the transmission robotic arm.

3. The system according to claim 2, characterized in that, The transmission robotic arm is equipped with a path planning module, which employs an improved search algorithm. The search algorithm obtains the currently available workspace information by querying the dynamic equipment topology map maintained by the central control unit, and integrates the 3D environmental point cloud data generated in real time by the image recognition unit to construct a cost map containing dynamic obstacles, and finally generates a collision-free and smooth motion trajectory.

4. The system according to claim 3, characterized in that, Before performing the measurement, the error measurement unit calls its built-in calibration module to execute a self-calibration program. The compensation coefficient used in the self-calibration program is obtained by the central control unit by looking up a table based on the historical temperature and humidity data reported by the environmental monitoring system, thereby eliminating system errors.

5. The system according to claim 4, characterized in that, The system supports hot-swapping of each functional unit. When the central control unit detects that a module has been removed or connected, it triggers a device re-enumeration process, updates the dynamic device topology map, and notifies the path planning module to recalculate all affected task paths.

6. The system according to claim 5, characterized in that, The central control unit also includes a data recording module for storing key parameters during all task execution processes. These key parameters are used for offline training and optimization of the deep reinforcement learning decision model.

7. The system according to claim 6, characterized in that, The task scheduling module calculates the overall system performance index in each decision cycle and uses this index to guide the optimization direction of the decision model. The formula for calculating the overall system performance index includes: Wherein, P is the overall system performance index; N is the total number of historical tasks recorded in the data recording module; and The pre-defined normalized weighting coefficients are used to balance the contributions of the two indicators. The ideal completion time for the k-th task, calculated based on task complexity, is obtained by the task parsing module of the central control unit. The actual completion time of the kth task recorded by the data recording module; The penalty factor, given by the system configuration parameters, is used to adjust the system's sensitivity and preference for completing tasks "timeout" or "early"; M is the total number of controllable modules in the system. and These are the maximum and minimum utilization rates of the i-th module within the statistical period, respectively, obtained from the statistics of the data recording module. A very small constant introduced to prevent division by zero errors.

8. The system according to claim 7, characterized in that, The ideal completion time is obtained by performing critical path analysis on the task using a directed acyclic graph. The calculation formula corresponding to the critical path analysis includes: Where Paths represents the set of all possible paths that constitute the task; p represents one of the paths; j represents a task node on path p; The standard operation time estimated by the task parsing module for the j-th task node; The baseline performance coefficients of the modules that process this task node are obtained by the central control unit from the module capability description file; and The complexity adjustment factor associated with the type of the j-th task node is given by the system experience parameter table; This refers to the offset between the current environmental parameters reported by the environmental monitoring system and the standard parameters.

9. The system according to claim 1, characterized in that, The image recognition unit integrates a depth camera and an RGB camera, and establishes a precise coordinate transformation relationship with the transmission robotic arm through a hand-eye calibration program.

10. The system according to claim 1, characterized in that, The insulation test chamber is equipped with a safety interlock mechanism that is directly connected to the central control unit, ensuring that the high-voltage test is only initiated when the chamber door is completely closed and the transmission robotic arm is in a safe position.

Images