A self-adaptive driving control method for a permanent magnet synchronous motor hand-push cloth cutting machine

Abstract

This invention discloses an adaptive drive control method for a manual push-type fabric cutting machine using a permanent magnet synchronous motor. The method uses the operating data of the permanent magnet synchronous motor as state features and the combination of motor control parameters as action parameters, establishing a mapping relationship. It designs a target reward function with optimal control strategy as a positive reward and knife jamming, wear, and high energy consumption as negative penalties. Using the state features as input, the method matches action parameters based on the target reward function, and the optimal decision as output, constructing a control parameter model for the manual push-type fabric cutting machine using a permanent magnet synchronous motor. The method collects the operating data of the permanent magnet synchronous motor in real time, inputs it into the control parameter model, and outputs the corresponding optimal decision. Based on the optimal decision, it adaptively adjusts the current loop and speed loop of the permanent magnet synchronous motor, driving the motor to dynamically adjust torque and speed, thus solving the problems of manually setting control parameters, inability to dynamically adapt to fabric types, and non-constant feed speeds.

Description

Technical Field

[0001] This invention belongs to the field of intelligent motor control technology, specifically relating to an adaptive drive control method for a permanent magnet synchronous motor hand-push fabric cutting machine, applicable to vertical fabric cutting machines in industries such as clothing and home textiles where manual feeding and up-and-down reciprocating sliding of the cutter are used. Background Technology

[0002] Currently, the motor drive control of hand-push fabric cutting machines is all fixed parameter open-loop control, which only sets a single fixed speed for the motor that slides the cutter up and down. This results in poor adaptability and no intelligent adaptive capability. The specific implementation method and its disadvantages are as follows:

[0003] 1. Operators judge the material, hardness, and thickness of the fabric to be cut based on their own experience, and manually set only the fixed operating speed of the motor on the equipment. Frontline operators are prone to misjudging the characteristics of the fabric, and the set speed does not match the actual cutting resistance. When the speed is too high, the blade will jam or break if the feed is too fast. When the speed is too low, the blade will repeatedly cut the fabric, causing fuzzing and uneven cuts. The accuracy deviation between the upper and lower layers of multi-layer cutting is large, and the fabric scrap rate is high.

[0004] 2. The equipment drives a permanent magnet synchronous motor at a fixed speed set manually. The cutter slides up and down at the same speed throughout the entire cutting process, without any dynamic adjustment. It completely ignores the changes in the feed speed caused by manual pushing. A few advanced machines have a pre-stored fixed speed mapping table for a limited number of fabrics, but manual selection of the fabric type is still required to match the preset speed. The fixed speed can only adapt to a single or a few types of fabrics. The cutting effect is poor when dealing with non-standard or composite fabrics. Furthermore, the cutter sharpening is triggered by manual timing or a fixed number of cuts, which cannot cope with the irregular fluctuations in the feed speed caused by manual pushing. The cutting stability is low, there is no quantitative cutting resistance data to support it, and it is impossible to accurately judge the dulling state. There is no control to differentiate between cutting and idle times when the cutter slides up and down. During idle times, it still runs at a fixed speed, causing motor idling losses. It cannot adjust the motor output according to the changes in cutting resistance during the cutting process. Sudden increases / decreases in resistance caused by changes in feed speed will lead to cutting quality problems, and the motor is prone to overload and overheating.

[0005] 3. The current, torque and other data of the motor are only used as the basis for the protection of the hardware overload trip, and are not used for the correction of drive parameters. When the manual feed knife is fast, the cutting resistance increases sharply and when the feed knife is slow, the cutting resistance drops sharply, and the system does not respond. When changing the fabric, the machine must be stopped and the speed must be reset, which takes a long time and reduces the overall production efficiency. Summary of the Invention

[0006] The technical problem that the present invention aims to solve is that existing hand-push fabric cutting machines rely on manual setting of control parameters, cannot dynamically adapt to fabric types and non-constant feed speeds, and cannot automatically set the rotation speed when changing fabrics.

[0007] To address the aforementioned technical problems, the present invention provides an adaptive drive control method for a permanent magnet synchronous motor-driven hand-push fabric cutting machine, comprising the following steps: The permanent magnet synchronous motor operation data reflecting the cutting conditions during the cutting process is selected as the state feature, and the combination of motor control parameters that control the motor torque and speed response characteristics is selected as the action parameter, and a mapping relationship is established with all the features in the state feature. Based on the relevant formulas of the risk coefficient of tool jamming, the wear coefficient of fabric, and the energy consumption coefficient of motor, the optimal control strategy is used as the positive reward, and tool jamming, wear, and high energy consumption are used as negative penalties, and a target reward function is designed. By taking state features as input, matching action parameters based on the objective reward function, and taking the optimal decision as output, a control parameter model for a permanent magnet synchronous motor push-type fabric cutting machine is constructed. The system collects real-time operating data of the permanent magnet synchronous motor, inputs it into the control parameter model of the permanent magnet synchronous motor hand-push fabric cutting machine, outputs the corresponding optimal decision, and adaptively adjusts the current loop and speed loop of the permanent magnet synchronous motor based on the optimal decision, driving the motor to dynamically adjust torque and speed.

[0008] Preferably, the state characteristic includes the current rise slope reflecting the fabric stiffness characteristics. The average q-axis current, reflecting the average cutting resistance. q-axis current variance, reflecting the degree of fluctuation in cutting resistance The average actual speed of the motor, reflecting its actual operating state. Motor speed variance, reflecting the degree of feed rate fluctuation. Speed ​​deviation rate, reflecting speed adaptability .

[0009] Preferably, the operating parameters include a q-axis current reference value for controlling the motor torque. The proportional coefficient of the speed loop PI controller used to control the speed response characteristics .

[0010] Preferably, the proportional coefficient of the speed loop PI regulator is... It adapts to the cutting characteristics of soft, medium-soft, medium-hard, hard, composite, and general unknown fabrics.

[0011] Preferably, after the current loop and speed loop of the permanent magnet synchronous motor are adaptively adjusted based on the optimal decision, the voltage reference value is converted into a voltage signal adapted to the motor drive through Clark transformation, Park transformation, and inverse Park transformation in sequence. Then, a complementary PWM drive signal is generated through the SVPWM algorithm to drive the motor to dynamically adjust the torque and speed.

[0012] Preferably, after the drive motor dynamically adjusts its torque and speed, the method further includes: calculating the reward value in real time based on the target reward function according to the cutting conditions after execution, and constructing the condition data into an experience sample and storing it in the experience pool to train the control parameter model of the permanent magnet synchronous motor push-type fabric cutting machine.

[0013] This invention provides an adaptive drive control method for a manual push-type fabric cutting machine using a permanent magnet synchronous motor. The method uses the operating data of the permanent magnet synchronous motor as state features and the combination of motor control parameters as action parameters, establishing a mapping relationship. Based on formulas for the risk coefficient of blade jamming, the fabric wear coefficient, and the motor energy consumption coefficient, an optimal control strategy is used as a positive reward, while blade jamming, wear, and high energy consumption are used as negative penalties, designing a target reward function. The state features are used as input, and action parameters are matched based on the target reward function, with the optimal decision as the output, constructing a control parameter model for the manual push-type fabric cutting machine using a permanent magnet synchronous motor. The operating data of the permanent magnet synchronous motor is collected in real time and input into the control parameter model, outputting the corresponding optimal decision. Based on the optimal decision, the current loop and speed loop of the permanent magnet synchronous motor are adaptively adjusted, driving the motor to dynamically adjust torque and speed, solving the problems of relying on manually set control parameters and the inability to dynamically adapt to fabric types and non-constant feed speeds in manual push-type cutting. Attached Figure Description

[0014] Figure 1 A schematic diagram of the hierarchical training process for the control parameter model of a permanent magnet synchronous motor-driven fabric cutting machine; Figure 2 This is a schematic diagram illustrating the adaptive adjustment process of the current loop and speed loop of a permanent magnet synchronous motor based on optimal decision-making. Figure 3 This is a schematic diagram illustrating the usage process of an adaptive drive control method for a hand-push fabric cutting machine with a permanent magnet synchronous motor. Detailed Implementation

[0015] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0016] An adaptive drive control method for a hand-push fabric cutting machine with a permanent magnet synchronous motor, provided by an embodiment of the present invention, includes the following steps: The operating data of the permanent magnet synchronous motor that can directly reflect the cutting conditions during the cutting process are selected as the state feature S.

[0017] The state feature S is normalized to the [0,1] interval to eliminate dimensional differences.

[0018] State characteristics S include: Current rise slope reflects fabric properties: This reflects the fabric's stiffness characteristics. The average q-axis current, reflecting the average cutting resistance. q-axis current variance, reflecting the degree of fluctuation in cutting resistance .

[0019] Reflecting cutting conditions: The average actual motor speed, reflecting the actual operating state of the motor. Motor speed variance, reflecting the degree of feed rate fluctuation. Speed ​​deviation rate, reflecting speed adaptability (The ratio of actual speed to reference speed).

[0020] A combination of motor control parameters that control the motor torque and speed response characteristics is selected as the action parameters, and a one-to-one mapping relationship is established with all features in the state feature S.

[0021] The operating parameters include: the q-axis current reference value used to control the motor torque. The proportional coefficient of the speed loop PI controller used to control the speed response characteristics .

[0022] Speed ​​ring PI regulator proportional coefficient It adapts to the cutting characteristics of soft, medium-soft, medium-hard, hard, composite, and general unknown fabrics.

[0023] Based on the relevant formulas of the risk coefficient of tool jamming, the wear coefficient of fabric, and the energy consumption coefficient of motor, a target reward function is designed with the optimal control strategy as the positive reward and tool jamming, wear, and high energy consumption as the negative penalties.

[0024] The weighting coefficients are set to α=0.5, β=0.3, and γ=0.2 respectively, and the reward value range is [-5,1].

[0025] Based on the average q-axis current The risk coefficient of the chuck is calculated by comparing the ratio of the current to the motor overload protection current threshold.

[0026] The fabric wear coefficient is calculated based on the difference between the actual rotational speed and the reference rotational speed, and the ratio of the difference to the reference rotational speed.

[0027] The motor energy consumption coefficient is calculated based on the ratio of the actual motor power to the rated motor power.

[0028] This invention constructs the core evaluation index of the reward function based on 6-dimensional state features. The specific calculation method of each coefficient is as follows: 1. Tool jamming risk factor: based on the average q-axis current. Motor overload protection current threshold The ratio is calculated using the following formula:

[0029] This coefficient directly reflects the degree of matching between motor torque and cutting resistance. At that time, it is judged as a high-risk situation for knife blocking, triggering a strong penalty.

[0030] 2. Fabric abrasion coefficient: based on actual rotation speed Compared with the reference speed The difference is calculated by combining the ratio of the difference to the reference speed, using the following formula:

[0031] This coefficient characterizes the degree of wear on the fabric caused by rotational speed fluctuations. If the fabric wear exceeds the limit, a penalty will be triggered.

[0032] 3. Motor energy consumption coefficient: based on the actual motor power. With the rated power of the motor The ratio is calculated using the following formula:

[0033] This coefficient is used to evaluate the energy efficiency level of motor operation, and to achieve synergistic optimization of cutting quality and energy consumption.

[0034] The slope of the current rise, which reflects the fabric's characteristics. q-axis current mean q-axis current variance And the average actual motor speed reflecting the cutting conditions. Motor speed variance Speed ​​deviation rate Using all 6-dimensional state features as input, and matching the optimal action parameters based on the objective reward function, a control parameter model for a permanent magnet synchronous motor push-type fabric cutting machine is constructed.

[0035] By taking state features as input, matching action parameters based on the objective reward function, and taking the optimal decision as output, a control parameter model for a permanent magnet synchronous motor push-type fabric cutting machine is constructed.

[0036] like Figure 1 As shown, the control parameter model of the permanent magnet synchronous motor hand-push fabric cutting machine is a forward propagation network, including: The input layer acquires state features based on the same number of neurons as the state features.

[0037] A single hidden convolutional layer extracts key information from state features, reducing interference from invalid features. The kernel size of the single hidden convolutional layer is 1×3, and the number of output channels is 8.

[0038] The fully connected layer establishes a non-linear mapping from state features to Q-values ​​in action parameters based on the mapping relationship.

[0039] The output layer selects the action parameter combination with the largest Q value as the optimal decision based on the objective reward function and the ε-greedy policy decision rule. At the same time, the output layer quantizes all weights and bias parameters from 32-bit floating-point to 8-bit signed integers (quantization range [-128, 127]), with a quantization error ≤5%, reducing the number of parameters by more than 75%, significantly reducing the computing power and storage requirements of model inference, and making it suitable for industrial embedded controllers.

[0040] The ε-greedy strategy decision-making rules are as follows: The initial exploration rate ε = 0.8 and the exploration rate decay coefficient η = 0.995 are set. The exploration rate gradually decreases with the increase of model training steps and eventually stabilizes at 0.1 to avoid the model from over-exploring in the later stage and causing decision instability. A random number ξ between 0 and 1 is randomly generated. If ξ < ε, the model randomly selects a set of control parameter combinations from the action space (exploring new working conditions and new strategies). If ξ ≥ ε, the model selects the control parameter combination with the largest output Q value (using the learned optimal strategy). The model inference time is ≤ 50ms, which matches the response speed of the motor closed-loop control and has no decision delay affecting the trimming process.

[0041] Real-time acquisition of operating data from permanent magnet synchronous motors is input into the control parameter model of the permanent magnet synchronous motor hand-push fabric cutting machine, and the corresponding optimal decision is output.

[0042] The real-time acquisition of operating data of the permanent magnet synchronous motor includes the following steps: With a fixed acquisition window of 200ms, the q-axis current and actual speed data of the permanent magnet synchronous motor are acquired synchronously. The acquisition frequency matches the motor's closed-loop control frequency, with no data delay, and is used as real-time acquisition data of the permanent magnet synchronous motor's operation.

[0043] Both single-hidden convolutional layers and fully connected layers use the ReLU activation function.

[0044] like Figure 2 As shown, the current loop and speed loop of the permanent magnet synchronous motor are adaptively adjusted based on the optimal decision. The voltage reference value after dual closed-loop regulation is converted into a voltage signal adapted to the motor drive through Clark transformation, Park transformation and inverse Park transformation in sequence. Then, a complementary PWM drive signal is generated through the SVPWM algorithm to drive the motor to dynamically adjust the torque and speed.

[0045] The motor output torque is matched with the actual cutting resistance in real time, which avoids tool jamming caused by sudden increase in resistance. It is suitable for the non-constant feed speed of manual pushing and achieves smooth speed adjustment under the premise of torque matching, avoiding excessive speed fluctuations that cause fabric wear. It is suitable for cutting conditions and replaces manual fixed speed setting.

[0046] Based on the cutting conditions after execution, the reward value is calculated in real time, and the condition data is used to build experience samples and stored in the experience pool to provide data support for subsequent model updates and optimizations.

[0047] Empirical Sample Construction: State Vector at the Next Moment After Parameter Execution , the current state The selected action Instant rewards Next state Quadruples are used to construct empirical samples; Sample storage: Experience samples are stored in an experience pool with a circular buffer structure. The capacity of the experience pool is set to 256 groups. When the sample size is ≥64 groups, 32 groups of samples are randomly sampled for model update.

[0048] When the sample size in the experience pool reaches a preset threshold, a dual-network structure is used to update the model offline, optimize model parameters, improve the accuracy of model decisions, and avoid the impact of sample correlation on training results. The specific update steps are as follows: Sample sampling: When the number of samples in the experience pool is ≥128, 32 samples are randomly sampled from the experience pool as the training set to avoid overfitting of the model due to the correlation of continuous samples. Dual-network definition: Set up an evaluation network and a target network with completely identical structures. The evaluation network is used to calculate the Q value of the current state, and the target network is used to calculate the target Q value of the next state. Q-value calculation: The network is evaluated based on the current state of the sampled data. Calculate the Q value of each action. ; The target network is based on the next state of the sampled sample. Calculate the optimal Q value ; Target Q-value calculation: combined with immediate rewards Calculate the target Q value using a discount factor γ (γ=0.9). The discount factor reflects the weight of future rewards in the model's decision-making. Parameter update: using the mean squared error loss function The error between the model's predicted value and the target value is calculated, and the weights and bias parameters of the evaluation network are updated using the stochastic gradient descent algorithm. Target network synchronization: Every 100 steps of model update, the parameters of the evaluation network are directly copied to the target network to ensure the stability of the model training process and avoid excessive fluctuations in the target value.

[0049] In the initial stage, the model mainly focuses on exploration, gradually learning the cutting characteristics of different fabrics and different feed speeds, optimizing control strategies, and replacing manual experience-based judgment. As the number of training steps increases, the exploration rate gradually decreases, the model mainly utilizes the learned optimal strategy, the decision accuracy continues to improve, and the adaptability of motor torque and speed is continuously optimized. When the equipment encounters old working conditions such as tool wear and motor performance degradation, or new working conditions such as cutting non-standard or composite fabrics, the model relearns the working condition characteristics through a small amount of exploration and adaptively adjusts the control parameters without the need for manual shutdown to set the speed.

[0050] This embodiment uses a 2kW permanent magnet synchronous motor hand-push type up-and-down sliding blade fabric cutting machine as the implementation object. This equipment is an existing conventional model that only supports manual setting of a fixed speed. It can cut six common types of fabrics, including cotton, linen, denim, leather, chiffon, and composite fabrics. The implementation process of this invention is described in detail. Colleagues can directly replicate this embodiment without additional professional equipment and technology.

[0051] Preparations before implementation Model training environment: The PC should be equipped with Python 3.8 or above, and TensorFlow Lite framework (dedicated to lightweight model training). A basic CPU is sufficient; a high-end graphics card is not required. Model training data: In a laboratory environment, cutting data of 6 types of fabrics were collected, with 200 sets of data collected for each type of fabric, for a total of 1200 sets. Each set of data includes 6-dimensional motor operating status features and corresponding cutting effect labels. Model training: A lightweight DQN model was built based on the TensorFlow Lite framework. The network structure was set according to the technical features of this invention. The model was trained using 1200 sets of collected data, iterated for 10000 steps, and after training, the model parameters were quantized into 8-bit integers and exported as a model file that can be recognized by the embedded controller. Equipment deployment: The quantized model file is burned into the industrial-grade embedded controller of the fabric cutting machine, and the control program of this invention is burned into it at the same time, including modules such as data acquisition, state construction, model inference, dual closed-loop control, sample storage, and model update, without the need to modify the equipment hardware; Parameter presets: The control program presets the experience pool capacity to 256 sets, the sampling batch to 32 sets, the initial exploration rate ε=0.8, the exploration rate decay coefficient η=0.995, the discount factor γ=0.9, and the target network update step size to 100 steps.

[0052] Core implementation steps (taking the cutting of 5mm thick stiff denim as an example, the existing equipment is prone to blade jamming when the rotation speed of the fabric is manually set). Equipment startup: Turn on the main power of the fabric cutting machine. The embedded controller completes initialization, including model loading, dual closed-loop control parameter initialization, and experience pool initialization. The equipment enters the cutting state without the need for manual setting of any speed. State vector construction: The operator pushes the fabric cutting machine to start cutting. The controller collects motor q-axis current and speed data in a 200ms acquisition window, and calculates 6-dimensional state feature values: , , , , , 5 (all values ​​are normalized), construct the state vector ; Model action decision: The controller randomly generates random numbers. ( After decay The model selects action 5, which outputs the largest Q value, and the corresponding control parameters are: , ; Dual closed-loop control execution: The above parameters are input into the motor dual closed-loop control module. The current loop uses 9A as the target value of the q-axis current and outputs a large torque to match the high cutting resistance of denim. The speed loop uses a proportional coefficient of 0.4 to slowly adjust the speed, adapt to the speed fluctuation of the hand-push knife, completely avoid the knife jamming, and replace the fixed speed of 3200rpm set by the manual for the fabric. Reward value calculation: Real-time collection of motor operation data yields the following calculated values: tool jamming risk coefficient = 0.85, fabric wear coefficient = 0.08, and motor energy consumption coefficient = 0.7. Substituting these values ​​into the reward function... This is a positive reward; Experience sample storage: Collect the state vector for the next time step. Constructing experience samples ( Action 5 , Store it in the experience pool; Model update and optimization: When the sample size in the experience pool reaches 128 groups, 32 groups of samples are randomly sampled, and the model parameters are updated through a dual network structure to optimize the prediction accuracy of the action Q value; Cyclic cutting: Repeat steps 2-7. The model continuously learns the cutting conditions of denim, and the accuracy of decision-making gradually improves. The cutting process is free from blade jamming and fabric fraying, and the cut is smooth. Compared with the existing method of manually setting a fixed speed, the cutting quality is greatly improved.

[0053] Example of cutting soft chiffon fabric (existing equipment with manually set rotation speed for this fabric is prone to wear). The operator pushes the fabric cutting machine to cut 0.1mm thick chiffon fabric, and the controller collects data to construct a state vector. ; The model selects action 1, and the corresponding control parameters are: , ; The current loop outputs a small torque to match the low cutting resistance of chiffon, and the speed loop uses a proportional coefficient of 0.6 to quickly adjust the rotation speed to match the feed rate, avoiding repeated cutting of the fabric by the tool; Calculate reward value As a result of high positive rewards, the model continuously optimizes the control strategy for the fabric, resulting in no pilling or offset during the cutting process. Compared to the existing method of manually setting a fixed speed of 3500 rpm, the fabric wear rate is reduced by 90%.

[0054] like Figure 3 As shown, the control method of this invention is integrated into the industrial-grade embedded controller of the fabric cutting machine, and the entire process is automated. Addressing the limitation of existing equipment that only allows manual speed setting, this invention completely eliminates the manual speed setting step. The operation process is intuitive and easy to understand, conforming to the habits of frontline operators, requiring no professional knowledge of motor control and intelligent algorithms. The specific usage steps are as follows: Equipment startup: Turn on the main power switch of the fabric cutting machine. The embedded controller will automatically complete the system initialization, including loading the lightweight DQN model, initializing the dual closed-loop control parameters of the permanent magnet synchronous motor, and initializing the experience pool. After initialization, the equipment indicator light will be constantly on, and the machine will enter the cutting state. If initialization fails, the indicator light will flash. After troubleshooting the controller, restart it. No manual setting of any speed is required. Cutting preparation: The operator lays the fabric to be cut flat and fixed on the worktable of the fabric cutting machine, and adjusts the cutting path of the blade according to the cutting requirements. No fabric parameter selection or motor speed setting is required. Manual push cutting: The operator holds the push handle of the fabric cutting machine and pushes the fabric cutting machine normally along the preset cutting path. There is no need to deliberately control the feed speed. The controller will collect the motor operation data in real time, complete all links such as state vector construction, model reasoning, and parameter execution, automatically adapt to the fluctuation of feed speed and fabric characteristics, and dynamically adjust the motor torque and speed, replacing the manual fixed speed setting. Real-time operating condition monitoring: The control panel of the fabric cutting machine will display the current motor q-axis current, actual speed, and cutting status (normal / warning) in real time. If there is a risk of blade jamming or excessive fabric wear, the panel will issue an audible and visual warning. The operator only needs to pause briefly, and the controller will automatically adjust the control parameters without stopping the machine to set the speed. Batch switching: When changing to fabrics of different materials, thicknesses, and layers, there is no need to turn off the power, adjust the speed, or set any parameters. Simply fix the new fabric on the worktable and push the cutting machine to start cutting. The controller's lightweight DQN model will automatically sense the characteristics of the new fabric and output the optimal control parameters after a small amount of exploration. The controller collects motor operating data in real time and calculates six dimensions of state features in sequence: current rise slope KIq, q-axis current mean Iqavg, q-axis current variance Iqvar, speed mean Navg, speed variance Nvar, and speed deviation rate Nerr. The lightweight DQN model is based on full state input and completes action decisions through an ε-greedy strategy. It combines a reward function to complete the working condition adaptation evaluation. After a small amount of online exploration and iteration, it automatically matches and outputs the corresponding optimal torque reference and speed loop control parameters. Achieve seamless switching and significantly reduce downtime for adjustments; Equipment maintenance: When the blades are severely worn or the motor malfunctions, the hardware protection module of the fabric cutting machine will trigger a shutdown. After the operator replaces the blades and repairs the motor, the machine can be restarted. The model will automatically adapt to the cutting conditions after the blades are replaced, without the need for retraining or manual setting of the speed. Equipment shutdown: Once all cutting tasks are completed, simply turn off the main power switch of the fabric cutting machine. The controller will automatically store the current model parameters and experience pool samples, which can be directly loaded the next time the machine is turned on. The control strategies already learned by the model will not be lost, and there is no need to retrain. Cutting can be started directly upon startup.

[0055] This invention, through dual-loop monitoring and coordinated control of the motor current loop and speed loop, adapts to the non-constant feed speed of manual push cutting, dynamically adjusting the speed and output torque of the drive motor for the up-and-down sliding of the cutter. This solves the problems of blade jamming, blade breakage, and uneven cuts caused by uneven feed speed during manual fabric cutting. At the same time, it achieves intelligent blade dulling recognition and overload safety protection, improving cutting quality and efficiency, extending the service life of the cutter and motor, without the need for additional sensors, resulting in low modification costs and suitability for the actual operating scenarios of manual push fabric cutting machines.

[0056] This invention establishes a correlation model between the current loop and cutting resistance. Based on the fluctuations in cutting resistance caused by changes in the hand-push knife speed, it dynamically adjusts the motor speed and torque of the up-and-down sliding knife to solve the problems of knife jamming and breakage. It achieves intelligent identification of the cutting stage and the idle stage of the knife, and adjusts the motor speed differently to reduce idle loss and reduce the difficulty of manual operation. There is no need to precisely control the hand-push knife speed. The system automatically adapts to the knife feed speed, reducing the labor intensity of operators. It is a pure algorithm-level improvement, which does not require modification of the fabric cutting machine's mechanical structure or the addition of additional sensors, reducing the cost of equipment upgrades and modifications. It is compatible with all existing hand-push up-and-down sliding knife fabric cutting machines.

[0057] This invention constructs a deep reinforcement learning framework adapted to fabric cutting conditions. It uses the operating data of a permanent magnet synchronous motor (PMSM) during the cutting process as the state space, the combination of motor control parameters as the action space, and designs a reward function with cutting stability and low loss as objectives. Through real-time interaction between a lightweight deep Q-network (DQN) model and the cutting conditions, it completes self-learning training, automatically outputting optimal control parameters to drive the current-speed dual closed-loop control module of the PMSM to dynamically adjust torque and speed. This invention requires no manual intervention in parameter setting, can adapt to different fabric characteristics and feed speed fluctuations, significantly reducing the jamming rate and fabric scrap rate. Furthermore, the lightweight model design is compatible with the computing power of industrial-grade embedded controllers, making it highly practical for engineering applications.

[0058] The technical effects of the embodiments of the present invention are as follows: 1. Eliminating the manual parameter setting process, the system autonomously senses the cutting conditions and fabric characteristics by collecting motor operation data, achieving identification of fabric type and cutting resistance without manual intervention.

[0059] 2. Construct a deep reinforcement learning framework that adapts to fabric cutting conditions. Through real-time interactive self-learning between the model and the cutting conditions, dynamically output the optimal motor control parameters to adapt to different fabric characteristics and non-constant feed speeds of manual pushing.

[0060] 3. Achieve autonomous optimization of control strategies, enabling the equipment to adapt to aging conditions such as tool wear and motor performance degradation, as well as the cutting needs of non-standard and composite fabrics, thereby improving the equipment's versatility and service life.

[0061] 4. Lightweight modifications are made to the deep reinforcement learning model to adapt it to the computing power and storage requirements of industrial-grade embedded controllers without increasing hardware costs, making it highly practical for engineering applications.

[0062] 5. Reduce manual operation intensity, improve the intelligence level of the fabric cutting machine, reduce downtime for adjustment, and improve production efficiency while ensuring cutting quality.

[0063] This invention is based on deep reinforcement learning theory, the vector control principle of permanent magnet synchronous motors, and the actual cutting conditions of a hand-push fabric cutting machine. All technical features are designed with clear principles and engineering basis. Targeted optimizations are made to address the limitation of existing equipment that can only manually set a fixed speed, ensuring the scientific validity and feasibility of the method, as detailed below: The application principle of deep reinforcement learning: Deep reinforcement learning acquires experience through real-time interaction between the agent and the environment, and optimizes decision-making strategies with the reward function as the target. It is suitable for control scenarios such as hand-push fabric cutting machines where "working conditions change dynamically, there is no precise mathematical model, and only a fixed speed can be set manually". It can achieve adaptive control of complex working conditions through self-learning, completely replacing manual speed judgment and setting. The design basis of the state space is that different fabrics have different physical properties (hardness, thickness, fiber density), and the cutting resistance during cutting varies significantly. This difference is directly reflected in the q-axis current (torque) and actual speed data of the permanent magnet synchronous motor. By using the statistical characteristics of current and speed (mean, variance, slope, etc.), the fabric characteristics and feed speed fluctuations can be effectively sensed. There is no need to install additional fabric detection sensors or make manual judgments, which meets the requirements of engineering practicality. The design basis of the motion space: Existing equipment can only set a fixed speed and cannot respond to changes in resistance. This invention directly controls the motor torque (matching the cutting resistance) through the q-axis current and adjusts the speed response characteristics through the speed loop PI parameters, replacing the manual fixed speed setting. Discrete parameter combinations are designed for the cutting resistance characteristics of different fabrics, which not only solves the adaptability problem of fixed speed, but also avoids the high computing power requirements of continuous motion space and adapts to the computing power limitations of industrial embedded controllers. The design basis of the reward function is as follows: The reward function takes the core production needs of the fabric cutting machine (stable cutting, low wear, and low energy consumption) as the goal. It guides the model to learn the optimal control strategy through positive rewards and negative penalties. The penalty conditions such as blade jamming and wear are the critical values ​​measured in engineering practice, which are in line with the quality judgment standards in actual production, so that the model decision is highly consistent with the actual production needs. The design basis of the lightweight DQN model: The control core of the push-type fabric cutting machine is an industrial-grade embedded controller with limited computing power and storage resources. By simplifying the structure (single convolutional layer + fewer neurons) and parameter quantization (8-bit integer), the computing power and storage requirements of the model can be greatly reduced. The model inference time is ≤100ms, which matches the response speed of the dual closed-loop control of the permanent magnet synchronous motor, and there is no decision delay affecting the cutting process. The design basis of the ε-greedy strategy: The ε-greedy strategy can balance the model's "exploration" and "utilization" capabilities. The high initial exploration rate ensures that the model can learn the characteristics of different fabrics and working conditions, replacing the accumulation of manual experience. The low exploration rate in the later stage ensures the stability of the model's decision-making and avoids fluctuations in cutting quality caused by over-exploration. The principle of dual closed-loop control for permanent magnet synchronous motors: The current loop, as the inner loop, has a fast response speed and can achieve precise torque control, matching cutting resistance in real time, thus solving the core problem that existing fixed-speed equipment cannot cope with changes in resistance; the speed loop, as the outer loop, has a slightly slower response speed and can dynamically adapt to fluctuations in feed speed. The combination of the two is a classic engineering solution for speed and torque control of permanent magnet synchronous motors, ensuring the stability and accuracy of motor control.

[0064] This invention addresses the limitation of existing equipment that can only manually set a fixed rotation speed by optimizing it. It is tailored to the actual production scenarios of hand-push fabric cutting machines, offering specific and quantifiable advantages. Compared to existing control methods, it has significant technical advantages, specifically: No manual speed setting is required, eliminating human judgment errors: The model autonomously senses the fabric characteristics and cutting conditions by collecting motor operation data, eliminating the need for manual judgment of the fabric and setting of the speed. The speed matching error rate is reduced from more than 15% to less than 2%, and the cutting qualification rate is increased from less than 85% to more than 98%, significantly reducing the fabric scrap rate. Real-time response to cutting resistance significantly improves cutting stability: The model dynamically outputs control parameters, and the drive motor torque is matched with the cutting resistance in real time, which avoids the problem of tool jamming from the root. At the same time, the speed is smoothly adjusted to adapt to the fluctuation of the feed speed, reducing the tool jamming rate by more than 35%, improving the cut flatness by more than 40%, and the accuracy deviation between the upper and lower layers of multi-layer cutting is ≤0.5mm, which solves the core defects of fixed speed equipment. Adaptive working conditions and adaptability to various types of fabrics: The model can dynamically adapt to various fabrics such as soft, medium-soft, medium-hard, hard, composite, and non-standard fabrics. There is no need to stop the machine to adjust parameters for different fabrics. It can still ensure cutting quality even when faced with irregular feed speed fluctuations caused by manual pushing, and the equipment's versatility is greatly improved. Self-learning and self-optimization capabilities, adapting to equipment aging and new working conditions: The model continuously optimizes the control strategy through real-time interaction with the cutting working conditions, and can automatically adapt to equipment aging conditions such as tool wear and motor performance degradation, without the need for manual readjustment, extending the equipment service life by more than 25%. Lightweight model design, strong engineering feasibility and low modification cost: After structural simplification and parameter quantification, the model can be deployed on various industrial-grade embedded controllers without the need for additional hardware computing modules. It can be achieved simply through software upgrades, with a modification cost of 0. It is suitable for upgrading existing fabric cutting machines without the need to replace equipment. Reduce manual labor intensity and improve production efficiency: Operators do not need to have professional knowledge of fabric judgment, nor do they need to frequently set and adjust the speed. They only need to push the cutting machine normally. Labor intensity is reduced by more than 70%, and no need to stop the machine for adjustment when changing fabric. The effective working time of the equipment is increased by more than 20%, and the overall production efficiency is increased by more than 15%. Multiple objectives are optimized to balance cutting quality and equipment economy: The reward function balances cutting stability, low fabric wear, and low motor energy consumption. While improving cutting quality, it can reduce motor energy consumption by more than 18%, reduce tool wear and motor overload, extend tool life by more than 30%, and motor life by more than 20%, thereby reducing the cost of equipment use and maintenance.

Claims

1. A self-adaptive driving control method for a permanent magnet synchronous motor hand-push cloth cutting machine, characterized in that, Includes the following steps: The permanent magnet synchronous motor operation data reflecting the cutting conditions during the cutting process is selected as the state feature, and the combination of motor control parameters that control the motor torque and speed response characteristics is selected as the action parameter, and a mapping relationship is established with all the features in the state feature. Based on the relevant formulas of the risk coefficient of tool jamming, the wear coefficient of fabric, and the energy consumption coefficient of motor, the optimal control strategy is used as the positive reward, and tool jamming, wear, and high energy consumption are used as negative penalties, and a target reward function is designed. By taking state features as input, matching action parameters based on the objective reward function, and taking the optimal decision as output, a control parameter model for a permanent magnet synchronous motor push-type fabric cutting machine is constructed. The system collects real-time operating data of the permanent magnet synchronous motor, inputs it into the control parameter model of the permanent magnet synchronous motor hand-push fabric cutting machine, outputs the corresponding optimal decision, and adaptively adjusts the current loop and speed loop of the permanent magnet synchronous motor based on the optimal decision, driving the motor to dynamically adjust torque and speed.

2. The adaptive driving control method for the permanent magnet synchronous motor hand-push type cloth cutting machine according to claim 1, characterized in that, The state features include a current rise slope reflecting a cloth hardness characteristic , a q-axis current mean value reflecting an average cutting resistance , a q-axis current variance reflecting a cutting resistance fluctuation degree , a motor actual speed mean value reflecting a motor actual operation state , a motor speed variance reflecting a feed speed fluctuation degree , a speed deviation rate reflecting a speed adaptability .

3. The adaptive driving control method of a permanent magnet synchronous motor hand-push type cloth cutting machine according to claim 1, characterized in that, The action parameter includes a q-axis current reference value for controlling motor torque a speed loop PI regulator proportional coefficient for controlling a speed response characteristic .

4. The adaptive driving control method of a permanent magnet synchronous motor hand-push type cloth cutting machine according to claim 1, characterized in that, The speed loop PI regulator proportional coefficient Adapt cutting characteristics for soft, medium soft, medium hard, hard, composite, general unknown materials.

5. The adaptive driving control method of the permanent magnet synchronous motor hand-push type cloth cutting machine according to claim 1, characterized in that, After adaptively adjusting the current loop and speed loop of the permanent magnet synchronous motor based on optimal decision, the voltage reference value is sequentially converted into a voltage signal adapted to the motor drive through Clark transformation, Park transformation, and inverse Park transformation. Then, a complementary PWM drive signal is generated through the SVPWM algorithm to drive the motor to dynamically adjust torque and speed.

6. The adaptive drive control method for a permanent magnet synchronous motor hand-push fabric cutting machine as described in claim 1, characterized in that, After the drive motor dynamically adjusts its torque and speed, the method further includes: calculating the reward value in real time based on the target reward function according to the cutting conditions after execution, and constructing the condition data into an experience sample and storing it in the experience pool to train the control parameter model of the permanent magnet synchronous motor push-type fabric cutting machine.

Images