Electrode insertion state recognition and cooperative control method based on multi-mode perception

Abstract

The invention discloses an electrode insertion state recognition and cooperative control method based on multi-mode sensing, and relates to the technical field of robot control, and the method comprises the steps: controlling a robot to grab an electrode and move the electrode to a position above a reaction tank; positioning the jack in the reaction tank according to the image acquired by the visual identification system and adjusting the clamping posture of the robot; the robot is controlled to insert the electrode into the jack, and multi-dimensional physical signals are dynamically collected; judging whether the electrode is normally inserted or not according to the multi-dimensional physical signal; if yes, using a pre-trained control strategy to output an action adjustment parameter according to the currently collected multi-dimensional physical signal; and the robot is controlled according to the action adjusting parameters, so that the electrode is normally inserted into the jack. On the basis of insertion judgment and control based on multi-source signal fusion, state time sequence modeling, recognition classification and control optimization cooperation, state sensing, autonomous judgment and deviation correction control of the robot in a corrosion electrode insertion task can be achieved, and the electrode insertion accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of robot control technology, and in particular to a method for electrode insertion state recognition and cooperative control based on multimodal perception. Background Technology

[0002] Electrochemical corrosion testing is a core method for studying the service stability, corrosion rate, and protection mechanism of metallic materials. Common testing modes include open-circuit potential, polarization curves, and impedance spectroscopy, all of which rely on stable and reliable electrode insertion. In high-throughput corrosion testing platforms, robotic systems are widely used to replace manual labor in repetitive operations such as electrode gripping, assembly, insertion, positioning, and wiring, thereby improving experimental efficiency and standardization. However, despite the increasing maturity of robotic hardware systems, existing insertion control processes generally employ an "open-loop" strategy based on position or time control, lacking real-time perception and judgment of the assembly process status. This leads to frequent assembly failures during actual testing, such as incomplete insertion, electrode non-contact, jamming, misalignment, loosening, or sample misplacement, severely affecting the accuracy of test data and the automation stability of the experimental platform. Summary of the Invention

[0003] In view of this, embodiments of this application provide a method for electrode insertion state identification and collaborative control based on multimodal sensing, so as to accurately insert electrochemically corroded electrodes.

[0004] One aspect of this application provides a method for electrode insertion state recognition and collaborative control based on multimodal sensing, the method comprising the following steps:

[0005] Control the robot to grasp the electrode and move it above the reaction tank;

[0006] The robot locates the socket in the reaction tank based on the image acquired by the visual recognition system and adjusts the gripping posture of the robot.

[0007] The robot is controlled to insert the electrode into the socket and dynamically acquire multidimensional physical signals;

[0008] The multidimensional physical signal is used to determine whether the electrode is properly inserted.

[0009] If the electrode is not properly inserted, the pre-trained control strategy is used to output action adjustment parameters based on the currently acquired multidimensional physical signals;

[0010] The robot is controlled according to the action adjustment parameters so that the electrode is properly inserted into the socket.

[0011] In some embodiments, the dynamic acquisition of multidimensional physical signals includes the following steps:

[0012] Using a visual recognition system, a six-dimensional force sensor, an electrochemical workstation, and a displacement sensor, the physical signals corresponding to the electrode during the insertion process are collected at a set frequency to form a multimodal sensing vector sequence. :

[0013] ;

[0014] in, For six-dimensional force information, To describe the homogeneous transformation matrix of position and orientation, This provides information on linear velocity and angular velocity. Visual pose error records the error between the current pose and the target pose. This refers to the circuit continuity status monitored by the electrochemical workstation.

[0015] In some embodiments, determining whether the electrode is properly inserted based on the multidimensional physical signal includes the following steps:

[0016] The multimodal sensing vector sequence is input into the state recognition model:

[0017] ;

[0018] Therefore, Bi-LSTM encoding is used to encode it as follows:

[0019] ;

[0020] ;

[0021] in, The bidirectional encoding features at time t. Forward LSTM sequence, The sequence is a backward LSTM sequence, trained using the Adam optimizer, with cross-entropy as the loss function;

[0022] Construct a state diagram with sample structure and action stage as nodes. Among them, nodes Represents the sensing state under each sensing channel, edge Indicates dynamic correlation;

[0023] The state graph is input into a graph neural network for processing, and the graph neural network outputs an overall graph embedding vector. This indicates the structural feature extraction result:

[0024] ;

[0025] The time representation extracted by Bi-LSTM Structure extracted by GNN The data is embedded and concatenated, then linearly transformed through a fully connected layer before being fed into a Softmax classifier.

[0026] ;

[0027] in, The classifier weight matrix is... For bias terms, The classification probability of a multi-class state;

[0028] The classification probability is used to determine whether the electrode is properly inserted.

[0029] In some embodiments, the pre-trained control strategy is a deep deterministic policy gradient algorithm;

[0030] The method of adjusting parameters by outputting actions based on the currently acquired multidimensional physical signals using a pre-trained control strategy includes the following steps:

[0031] The state vector is determined based on the currently acquired multidimensional physical signals and the classification probability. :

[0032] ;

[0033] The depth-deterministic policy gradient algorithm is used based on the state vector. Output motion adjustment parameters :

[0034] ;

[0035] in, This represents the policy function, i.e., the Actor network in the deep deterministic policy gradient algorithm; For the parameter set of the Actor network; Indicates the amount of insertion force adjustment; Indicates the amount of pose adjustment; Indicates the speed adjustment amount.

[0036] In some embodiments, the deep deterministic policy gradient algorithm includes an Actor network, a Critic network, and a reward function;

[0037] The Actor network is a multilayer perceptron with a three-layer structure. The input layer has a 20-dimensional dimension, the hidden layer has 256-128-64 units with ReLU activation, and the output layer uses tanh activation to normalize the action range. The input of the Actor network is... The output is the motion adjustment parameters. ;

[0038] The Critic network is an MLP structure with 256-128-64 hidden layers. The input to the Critic network is... The output is a Q-value estimate, including a single scalar and an evaluation of the quality of the action;

[0039] The reward function includes: successful insertion and conduction: +100; successful correction after anomaly detection: +50; constraint target for force / displacement per step: + ( ), where max_error is the preset threshold; failure retry exceeding the threshold: -50; jamming or damage: -100;

[0040] The deep deterministic policy gradient algorithm is trained through the following steps:

[0041] Offline training uses an experience replay buffer, where the buffer size is... Batch size 64; target network update, where =0.005; Noise exploration uses the Ornstein-Uhlenbeck process; Optimizer Adam, with a learning rate of 0.001 for the Actor network and 0.002 for the Critic network; Training objective minimizes the TD error, with a convergence criterion of policy variance <0.05; Parameters are updated every 10 actual instrumentation tasks, fine-tuned using real data.

[0042] In some embodiments, controlling the robot according to the action adjustment parameters to ensure that the electrode is properly inserted into the socket includes the following steps:

[0043] The robot's insertion force, pose, and speed are adjusted according to the motion adjustment parameters to ensure that the electrode is properly inserted into the socket.

[0044] In some embodiments, the method further includes the following steps:

[0045] If the electrode is properly inserted, a test start signal is sent to the electrochemical workstation to control the electrochemical workstation to perform the test.

[0046] Another aspect of this application embodiment provides an electrode insertion state recognition and collaborative control device based on multimodal sensing, the device comprising:

[0047] The electrode gripping unit is used to control the robot to grip the electrode and move it above the reaction tank;

[0048] The positioning and adjustment unit is used to locate the socket in the reaction tank and adjust the gripping posture of the robot based on the image obtained by the visual recognition system.

[0049] An insertion and sensing unit is used to control the robot to insert the electrode into the insertion hole and dynamically acquire multi-dimensional physical signals;

[0050] An insertion determination unit is used to determine whether the electrode is properly inserted based on the multidimensional physical signal.

[0051] The parameter adjustment unit is used to adjust the parameters by outputting an action based on the currently acquired multidimensional physical signal using a pre-trained control strategy if the electrode is not properly inserted.

[0052] The motion adjustment unit is used to control the robot according to the motion adjustment parameters so that the electrode is properly inserted into the socket.

[0053] Another aspect of this application embodiment provides an electronic device, including a processor and a memory;

[0054] The memory is used to store programs;

[0055] The processor executes the program to implement any of the methods described above.

[0056] Another aspect of this application provides a computer-readable storage medium storing a program that is executed by a processor to implement the method described in any of the above embodiments.

[0057] This application includes at least the following beneficial effects:

[0058] This application controls a robot to grasp an electrode and move it above a reaction tank; locates the insertion hole in the reaction tank based on images acquired by a vision recognition system and adjusts the robot's gripping posture; controls the robot to insert the electrode into the insertion hole while dynamically acquiring multi-dimensional physical signals; determines whether the electrode is correctly inserted based on the multi-dimensional physical signals; if the electrode is not correctly inserted, a pre-trained control strategy is used to output action adjustment parameters based on the currently acquired multi-dimensional physical signals; and the robot is controlled according to the action adjustment parameters to ensure that the electrode is correctly inserted into the insertion hole. This application, based on multi-source signal fusion, state-time modeling, and collaborative operation of recognition classification and control optimization, enables the robot to achieve state perception, autonomous judgment, and corrective control in actual electrode insertion tasks, thereby improving the accuracy of electrode insertion. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 A flowchart illustrating an electrode insertion state recognition and collaborative control method based on multimodal sensing, provided for an embodiment of this application;

[0061] Figure 2 This is a schematic diagram of the electrode co-insertion system provided in the embodiments of this application;

[0062] Figure 3 A schematic flowchart illustrating the electrode co-insertion method provided in an embodiment of this application;

[0063] Figure 4 This is a structural block diagram of an electrode insertion state recognition and collaborative control device based on multimodal sensing, provided in an embodiment of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] Before providing a detailed description of the embodiments of this application, some related technologies involved in the embodiments of this application will be described first, as follows:

[0066] In existing automated assembly systems, robots typically execute a pre-set sequence of actions, including electrode grasping, movement, alignment, insertion, and release. These actions are mostly based on trajectory planning or position templates, and the judgment criterion is limited to whether the robotic arm has reached the target pose, without sensing or analyzing dynamic feedback information during the insertion process. During insertion, actual changes in insertion force, the fit between the electrode and the hole, and whether the electrode gets stuck or shifted are not monitored or identified by the system. If the electrode is not fully inserted or has poor contact, the system will still default to "insertion completed" and directly start the electrochemical testing program, ultimately leading to an open circuit, signal drift, or measurement failure. Due to the lack of feedback and fault tolerance, such problems are often only discovered through post-event manual analysis of abnormal results, greatly limiting the practicality of high-throughput platforms.

[0067] To overcome these problems, some studies have attempted to introduce single-channel sensors, such as end-effector force sensors to measure clamping or insertion forces for a rough assessment of assembly quality, or position encoders to determine whether end-effector displacement has reached a preset threshold. Some systems have also incorporated visual recognition modules to detect sample position and orientation. However, single-sensor signals exhibit poor stability and resolution during insertion, especially under conditions of complex samples, tolerance interference, and liquid disturbance at the injection port. Single signals are easily susceptible to interference and misinterpretation. For example, displacement sensors alone cannot distinguish between "not fully inserted" and "stuck midway," force sensors alone cannot determine whether the closed electrode circuit has truly been contacted, and vision systems alone cannot accurately determine insertion depth and conduction status. The lack of multimodal fusion sensing and temporal modeling mechanisms remains a key bottleneck that current technologies have not yet resolved.

[0068] Furthermore, even if some systems can perform state recognition through hard-coded rules, such as setting "force threshold + position threshold" as judgment criteria, these static rules cannot adapt to complex and dynamically changing scenarios, nor can they form self-correction and optimization capabilities. Once insertion failure occurs, the robot can only simply retry the fixed action, lacking the ability to adjust the range of motion, path planning, and clamping force. Existing systems generally lack a control framework that combines state recognition with policy optimization, and the robot cannot dynamically adjust its behavior based on the recognition results to achieve adaptive action.

[0069] In summary, current automated electrochemical electrode insertion systems lack a complete state recognition and intelligent control mechanism, exhibiting key problems such as limited recognition methods, rigid control strategies, low assembly success rates, and unrecoverable faults. These issues severely restrict the development of corrosion testing platforms towards high efficiency, stability, and intelligence. Therefore, there is an urgent need for an insertion judgment and control method capable of multi-source signal fusion, state timing modeling, and coordinated operation of recognition classification and control optimization. This would enable robots to achieve state perception, autonomous judgment, and corrective control capabilities in actual corrosion electrode insertion tasks.

[0070] This application relates to the fields of electrochemical corrosion testing technology and intelligent robot control, and in particular to an electrode insertion state recognition and control method that combines multimodal perception and strategy optimization. It aims to solve key technical problems in current automated corrosion test platforms, such as the inability to judge electrode insertion quality in real time, the difficulty in effectively confirming contact status, and the lack of feedback error correction mechanism for robot motion rigidity.

[0071] To address the problems of existing automated systems' single-mode insertion recognition methods, reliance on a single sensor (such as force or displacement) leading to poor resolution and susceptibility to interference and misjudgment, and the inability of static state judgment rules (such as fixed thresholds) to adapt to dynamic changing scenarios, this application integrates multimodal dynamic data such as force, displacement, electrode conduction status, and visual positioning. It utilizes a bidirectional long short-term memory (Bi-LSTM) network to model the time-series signal and constructs a multi-class insertion state classifier by integrating spatial topology and inter-sample correlation information through a graph neural network (GNN). This can effectively improve the recognition accuracy to over 95% and reduce the range of anomaly detection errors.

[0072] To address the problems of rigid robot control strategies that rely on fixed instructions and open-loop execution, resulting in the inability to dynamically correct errors after insertion failures, low assembly success rates, and unrecoverable faults, this application proposes a closed-loop control mechanism based on Deep Deterministic Policy Gradient (DDPG). This mechanism uses state classification results as environmental feedback to dynamically adjust insertion force, end effector posture, insertion path, or execute retry actions. It integrates reward function design and online fine-tuning to achieve adaptive optimization, significantly reduce the average number of retries, and cope with complex working conditions.

[0073] To address the issues of existing systems lacking feedback and fault tolerance capabilities, and assembly problems only being discovered manually afterward, leading to significant deviations in experimental results and increased sample scrap rates, this application constructs a closed-loop collaborative mechanism of "perception-identification-decision-execution." Using the continuity of the electrochemical workstation circuit as the core criterion, it confirms the insertion quality in real time and triggers adjustments when anomalies occur, ensuring effective electrode continuity before testing and improving data reliability and automation levels.

[0074] Reference Figure 1 This application provides a method for electrode insertion state recognition and collaborative control based on multimodal sensing, specifically including the following steps S100~S150:

[0075] S100: Controls the robot to grasp the electrode and move it above the reaction tank;

[0076] S110: Locate the socket in the reaction tank based on the image obtained by the visual recognition system and adjust the clamping posture of the robot;

[0077] S120: Control the robot to insert the electrode into the socket and dynamically acquire multi-dimensional physical signals;

[0078] S130: Determine whether the electrode has been properly inserted based on the multidimensional physical signal;

[0079] S140: If the electrode is not properly inserted, the pre-trained control strategy is used to output action adjustment parameters based on the currently acquired multidimensional physical signal;

[0080] S150: Control the robot according to the action adjustment parameters so that the electrode is properly inserted into the socket.

[0081] The following section will provide a detailed introduction and explanation of the solutions in the embodiments of this application, using specific application examples.

[0082] Specifically, this embodiment includes the following technical solutions:

[0083] (1) Products or equipment.

[0084] This embodiment provides a method for electrode insertion state recognition and robot collaborative control based on multimodal perception and strategy optimization, aiming to solve the problem in automated electrochemical corrosion testing where robots cannot determine whether the insertion action is successful or whether the contact is closed. This embodiment uses a standard electrochemical testing system as the core platform, supplemented by a multi-degree-of-freedom robot, sensing units, and state recognition and control algorithms, to construct a closed-loop collaborative insertion control mechanism of "perception-recognition-decision-execution". This system can realize real-time judgment of the insertion state of corrosion samples, adjust actions based on the recognition results, ensure the insertion success rate and test connection rate of working electrodes, and improve the reliability and intelligence level of the automated corrosion testing platform.

[0085] ① Components necessary to constitute the product:

[0086] Reference Figure 2 The system comprises the following key components: an industrial robot, an end effector gripper module, a six-dimensional force sensor, a displacement encoder, a vision recognition system, an electrochemical workstation, and a control system host.

[0087] ②The configuration relationship between the various components:

[0088] The aforementioned components are integrated into the same platform with functional coupling and task flow sequence. The industrial robot serves as the core execution unit, connected to the end effector via a flange, and a six-dimensional force sensor is installed between the end effector and the body to achieve precise force feedback during the insertion process; the end effector module is designed with an electrode clamping interface to accommodate various standard sample sizes; the vision recognition system is fixed above the test platform to provide visual guidance before assembly and assist in subsequent status recognition; the top contact end of the working electrode and the electrical contact detection device are integrated in the upper part of the cell and connected to the electrochemical workstation test interface.

[0089] After completing sample insertion, the robot synchronously sends force, displacement, and electrical contact information collected by various sensors to the control system host for status determination. The identification results are sent to the control strategy module, where DDPG generates the current adjustment strategy and applies it to the robot system. If the status is determined to be "insertion completed and conduction successful," the system links with the electrochemical workstation to trigger the next experiment; otherwise, it enters the adjustment / retry process. The entire platform is coordinated by a task scheduling system that unifies the scheduling of command flow, sample flow, and data flow.

[0090] ③ The connection methods between the components (such as connection relationships):

[0091] In terms of hardware connectivity, the six-dimensional force sensor is integrated with the gripper and robot via a mechanical flange, and transmits the collected data to the control system host in real time via an industrial Ethernet interface (EtherCAT protocol). The displacement encoder directly interfaces with the robot's control system, providing end-effector position and attitude data to the control system host. The electrochemical workstation maintains information synchronization with the control system host via a network port, using open-circuit potential to detect electrode continuity. The vision recognition system transmits image frames with the control system host via a gigabit network interface. The control system host integrates unified task nodes through an internal bus, and the system dynamically adjusts the industrial robot's movements based on vision and circuit continuity status, ultimately controlling the electrochemical workstation to start testing via a TCP / IP interface.

[0092] (2) Working principle and action process (refer to) Figure 3 ).

[0093] ① Sample grabbing and insertion process:

[0094] The system calls a preset trajectory to control the robot to move to the target position according to the fixed arrangement coordinates of the sample tray, executes the gripper grasping action, completes the stable pickup of the electrode to be tested, and prepares to enter the insertion process.

[0095] ② Positioning and attitude adjustment before insertion:

[0096] The robot moves the electrode above the reaction tank, and the vision recognition system assists in positioning the insertion hole and adjusting the clamping posture to ensure that the insertion direction is vertical and the end is aligned with the center.

[0097] ③ Insertion process and sensor data acquisition:

[0098] The robot's end effector slowly presses down along the Z-axis to insert the electrode. A vision recognition system, a six-dimensional force sensor, an electrochemical workstation, and a displacement sensor collect multi-dimensional physical signals during the insertion process at high frequency, forming a sensing vector sequence. :

[0099] ;

[0100] in For six-dimensional force information, To describe the homogeneous transformation matrix of position and orientation, This provides information on linear velocity and angular velocity. Visual pose error records the error between the current pose and the target pose. This refers to the circuit continuity status monitored by the electrochemical workstation.

[0101] ④ Status recognition and judgment:

[0102] Input multimodal time series signals into the state recognition model Encoded using Bi-LSTM:

[0103] ;

[0104] ;

[0105] in The bidirectional encoded features at time t, with a hidden layer dimension of 128, are fed forward LSTM sequences. and backward LSTM sequences Each dimension is 64, trained using the Adam optimizer with a learning rate of 0.001 and a loss function of cross-entropy.

[0106] Construct a state diagram with sample structure and action stage as nodes. , where nodes Represents the sensing state under each sensing channel, edge This represents dynamic relationships (such as force-displacement coupling). The graph is fed into a graph neural network (GNN) for processing, outputting a global graph embedding vector. This indicates the structural feature extraction result:

[0107] ;

[0108] Finally, the system extracts the timing representation from Bi-LSTM. Structure extracted by GNN The data is embedded and concatenated, then linearly transformed through a fully connected layer before being fed into a Softmax classifier.

[0109] ;

[0110] in, This is the classifier weight matrix (dimension 5x64, where 5 represents the number of classes). For bias terms, Output the probability of classifying multiple states. The output labels include abnormalities such as: insertion completed and making good contact, insufficient insertion depth, jamming or skew, non-conductive contact, and sample structure damage.

[0111] ⑤ Control strategy adjustment and action correction:

[0112] If the connection fails or the status is identified as "not fully inserted," "skewed and stuck," or "poor contact," the system invokes the trained DDPG control strategy, based on the current status and the classifier's output label. as a state vector :

[0113] ;

[0114] Output motion adjustment parameters :

[0115] ;

[0116] In the formula This represents the policy function, i.e., the Actor network in the DDPG algorithm; The set of parameters (weights and biases) for this network is continuously updated during training to approximate the optimal policy; Indicates the amount of insertion force adjustment; Indicates the amount of pose adjustment; This indicates the speed adjustment amount. Based on this, the robot fine-tunes its insertion force, pose, speed, and other motion parameters, re-executes the insertion action, and enters a new round of identification and judgment. The system records the number of retries and sets a threshold (e.g., 3 times) to determine whether to enter the abnormal process.

[0117] ⑥ Assembly confirmation and test triggering:

[0118] Once the insertion process is complete and the working electrode is conductive, the control system sends a test start signal to the electrochemical workstation, which then performs polarization or impedance testing. The test data, along with the insertion process status, is written to the database. The system then proceeds to the next sample processing cycle.

[0119] (3) DDPG training process.

[0120] The DDPG architecture employs an actor-critic framework to optimize the continuous action space.

[0121] Actor Network: Multilayer Perceptron (MLP), with a 3-layer structure (20-dimensional input layer, 256-128-64 units in hidden layers, ReLU activation function, and tanh activation in the output layer to normalize the action range). Input Output continuous action vectors .

[0122] Critic Network: Similar to an MLP structure (hidden layers 256-128-64), input... Output Q-value estimate (single scalar, for evaluating action quality).

[0123] Reward function: Sparse + dense combination design. Successful insertion and conduction: +100; Successful correction after anomaly detection: +50; Constraint target for force / displacement per step: + ( The threshold value is set as follows: max_error is a preset threshold (e.g., force error of 10N); failure to retry exceeds the threshold: -50; jamming or damage: -100. Rewards encourage rapid convergence and safety (avoiding excessive force that could damage the sample).

[0124] Training process: Offline training uses an experience replay buffer (buffer size) Batch size 64); Target network update ( =0.005); noise exploration uses the Ornstein-Uhlenbeck process; optimizer Adam (learning rate Actor 0.001, Critic 0.002); training objective is to minimize TD error, with convergence criterion being policy variance <0.05. The model is updated every 10 actual instrumentation tasks, fine-tuned using real data.

[0125] Beneficial effects:

[0126] I. Significantly improved intelligent judgment and control precision of insertion actions. This embodiment integrates multiple sensing channels, including six-dimensional force sensing, end-effector pose, visual pose error, and electrochemical conduction signals, to construct a temporal state recognition model and a graph structure space analysis model. This enables accurate identification and classification of abnormal states such as "not fully inserted," "stuck," "skewed," and "not conductive" during the insertion process. The system can perceive changes in assembly quality in real time and generate status labels as the basis for subsequent action control decisions, breaking away from the static execution mode of "only executing, not judging" in traditional automatic insertion processes.

[0127] II. Adaptive optimization of robot motion and closed-loop correction control of the insertion process were achieved. This embodiment constructs a continuous motion output model based on the Deep Deterministic Policy Gradient (DDPG) algorithm, which can dynamically adjust the insertion force, end effector velocity, and insertion posture according to the current state vector, realizing fine-tuning and trial-and-error correction of the motion strategy. The robot no longer relies on a fixed trajectory, but can optimize the execution path in real time according to the actual state of each assembly, significantly improving the insertion success rate and compatibility with complex samples.

[0128] III. A closed-loop criterion for insertion results, centered on the state of the electrochemical workstation circuit, was constructed to ensure the authenticity and validity of the electrode conductivity state before testing. Compared with traditional completion judgment methods based on position or time, this embodiment focuses on "whether a stable electrochemical circuit can be established" as the criterion for determining insertion completion. This effectively avoids test errors caused by poor electrode contact, significantly improves data quality and the reliability of the testing system, and meets the engineering requirements of high-throughput automated corrosion platforms.

[0129] Reference Figure 4 This application provides an electrode insertion state recognition and collaborative control device based on multimodal sensing, comprising:

[0130] The electrode gripping unit is used to control the robot to grip the electrode and move it above the reaction tank;

[0131] The positioning and adjustment unit is used to locate the socket in the reaction tank and adjust the gripping posture of the robot based on the image acquired by the visual recognition system.

[0132] An insertion and sensing unit is used to control the robot to insert the electrode into the insertion hole and dynamically acquire multi-dimensional physical signals;

[0133] An insertion determination unit is used to determine whether the electrode is properly inserted based on the multidimensional physical signal.

[0134] The parameter adjustment unit is used to adjust the parameters by outputting an action based on the currently acquired multidimensional physical signal using a pre-trained control strategy if the electrode is not properly inserted.

[0135] The motion adjustment unit is used to control the robot according to the motion adjustment parameters so that the electrode is properly inserted into the socket.

[0136] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0137] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0138] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0139] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0140] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0141] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0142] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0143] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0144] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0145] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for electrode insertion state recognition and collaborative control based on multimodal sensing, characterized in that, The method includes the following steps: Control the robot to grasp the electrode and move it above the reaction tank; The robot locates the socket in the reaction tank based on the image acquired by the visual recognition system and adjusts the gripping posture of the robot. The robot is controlled to insert the electrode into the socket and dynamically acquire multidimensional physical signals; The multidimensional physical signal is used to determine whether the electrode is properly inserted. If the electrode is not properly inserted, the pre-trained control strategy is used to output action adjustment parameters based on the currently acquired multidimensional physical signals; The robot is controlled according to the action adjustment parameters so that the electrode is properly inserted into the socket.

2. The electrode insertion state recognition and collaborative control method based on multimodal sensing according to claim 1, characterized in that, The dynamic acquisition of multidimensional physical signals includes the following steps: Using a visual recognition system, a six-dimensional force sensor, an electrochemical workstation, and a displacement sensor, the physical signals corresponding to the electrode during the insertion process are collected at a set frequency to form a multimodal sensing vector sequence. : ； in, For six-dimensional force information, To describe the homogeneous transformation matrix of position and orientation, This provides information on linear velocity and angular velocity. Visual pose error records the error between the current pose and the target pose. This refers to the circuit continuity status monitored by the electrochemical workstation.

3. The electrode insertion state recognition and collaborative control method based on multimodal sensing according to claim 2, characterized in that, The step of determining whether the electrode is properly inserted based on the multidimensional physical signal includes the following steps: The multimodal sensing vector sequence is input into the state recognition model: ； Therefore, Bi-LSTM encoding is used to encode it as follows: ； ； in, The bidirectional encoding features at time t. Forward LSTM sequence, The sequence is a backward LSTM sequence, trained using the Adam optimizer, with cross-entropy as the loss function; Construct a state diagram with sample structure and action stage as nodes. Among them, nodes Represents the sensing state under each sensing channel, edge Indicates dynamic correlation; The state graph is input into a graph neural network for processing, and the graph neural network outputs an overall graph embedding vector. This indicates the structural feature extraction result: ； The time representation extracted by Bi-LSTM Structure extracted by GNN The data is embedded and concatenated, then linearly transformed through a fully connected layer before being fed into a Softmax classifier. ； in, The classifier weight matrix is... For bias terms, The classification probability of a multi-class state; The classification probability is used to determine whether the electrode is properly inserted.

4. The electrode insertion state recognition and collaborative control method based on multimodal sensing according to claim 3, characterized in that, The pre-trained control strategy is a deep deterministic policy gradient algorithm. The method of adjusting parameters by outputting actions based on the currently acquired multidimensional physical signals using a pre-trained control strategy includes the following steps: The state vector is determined based on the currently acquired multidimensional physical signals and the classification probability. : ； The depth-deterministic policy gradient algorithm is used based on the state vector. Output motion adjustment parameters : ； in, This represents the policy function, i.e., the Actor network in the deep deterministic policy gradient algorithm; For the parameter set of the Actor network; Indicates the amount of insertion force adjustment; Indicates the amount of pose adjustment; Indicates the speed adjustment amount.

5. The electrode insertion state recognition and collaborative control method based on multimodal sensing according to claim 4, characterized in that, The deep deterministic policy gradient algorithm includes an Actor network, a Critic network, and a reward function; The Actor network is a multilayer perceptron with a three-layer structure. The input layer has a 20-dimensional dimension, the hidden layer has 256-128-64 units with ReLU activation, and the output layer uses tanh activation to normalize the action range. The input of the Actor network is... The output is the motion adjustment parameters. ; The Critic network is an MLP structure with 256-128-64 hidden layers. The input to the Critic network is... The output is a Q-value estimate, including a single scalar and an evaluation of the quality of the action; The reward function includes: successful insertion and conduction: +100; successful correction after anomaly detection: +50; constraint target for force / displacement per step: + ( ), where max_error is the preset threshold; failure retry exceeding the threshold: -50; jamming or damage: -100; The deep deterministic policy gradient algorithm is trained through the following steps: Offline training uses an experience replay buffer, where the buffer size is... Batch size 64; target network update, where =0.005; Noise exploration uses the Ornstein-Uhlenbeck process; Optimizer Adam, with a learning rate of 0.001 for the Actor network and 0.002 for the Critic network; Training objective minimizes the TD error, with a convergence criterion of policy variance <0.05; Parameters are updated every 10 actual instrumentation tasks, fine-tuned using real data.

6. The electrode insertion state recognition and collaborative control method based on multimodal sensing according to claim 1, characterized in that, The step of controlling the robot by adjusting parameters according to the action to ensure that the electrode is properly inserted into the socket includes the following steps: The robot's insertion force, pose, and speed are adjusted according to the motion adjustment parameters to ensure that the electrode is properly inserted into the socket.

7. A method for electrode insertion state recognition and collaborative control based on multimodal sensing according to any one of claims 1 to 6, characterized in that, The method further includes the following steps: If the electrode is properly inserted, a test start signal is sent to the electrochemical workstation to control the electrochemical workstation to perform the test.

8. A device for electrode insertion state recognition and collaborative control based on multimodal sensing, characterized in that, The device includes: The electrode gripping unit is used to control the robot to grip the electrode and move it above the reaction tank; The positioning and adjustment unit is used to locate the socket in the reaction tank and adjust the gripping posture of the robot based on the image acquired by the visual recognition system. An insertion and sensing unit is used to control the robot to insert the electrode into the insertion hole and dynamically acquire multi-dimensional physical signals; An insertion determination unit is used to determine whether the electrode is properly inserted based on the multidimensional physical signal. The parameter adjustment unit is used to adjust the parameters by outputting an action based on the currently acquired multidimensional physical signal using a pre-trained control strategy if the electrode is not properly inserted. The motion adjustment unit is used to control the robot according to the motion adjustment parameters so that the electrode is properly inserted into the socket.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a program that is executed by a processor to implement the method as described in any one of claims 1 to 7.

Images