Intelligent winding and detection method and system for three-phase integrated coil of soft start device

Abstract

The present invention provides a method and system for intelligent winding and detection of a three-phase integrated coil of a soft start device, which relates to the technical field of three-phase integrated coil winding. The method includes: determining the electrical parameters of coil winding and preparing winding materials; calculating the optimal tension within k moments through establishing a tension mathematical model to adaptively control the wire tension; using a neural network model to perform real-time correction and compensation on key parameters; a vision detection control center detecting the coil in real time to determine whether errors occur during the winding process, and if so, stopping the winding and issuing an alarm; putting the wound coil into epoxy resin glue for degassing to ensure that the insulating material covers the surface of the wire and the gaps in the coil. By using a model predictive control algorithm and a neural network model to adaptively adjust the wire tension, the present invention controls the wire tension within a preset range, avoids damage to the wire due to excessive stretching or relaxation, helps to maintain the integrity and stability of the wire, and improves the overall performance of the soft start device.

Description

Technical Field

[0001] The present invention relates to the technical field of three-phase integration, and particularly to an intelligent winding and detection method for a three-phase integrated coil of a soft start device and an intelligent winding and detection system for a three-phase integrated coil of a soft start device. Background Art

[0002] The winding of the three-phase integrated coil of the soft start device refers to winding a wire into a three-phase integrated coil according to specific rules and parameters and applying it to a soft starter. This winding method enables the coil to generate a smooth magnetic field change when energized, thereby realizing the smooth start and speed control of the motor.

[0003] If the wire tension is not properly controlled, it may have a serious impact on the performance and service life of the soft start device. Excessive tension may cause problems such as coil deformation and insulation layer damage; while too little tension may cause problems such as coil looseness and deterioration of electrical performance. These problems will directly affect the normal operation and service life of the soft start device.

[0004] When the wire tension is too large, the coil may be subjected to excessive mechanical stress during the winding process, resulting in a change in the shape of the coil. This deformation may affect the mechanical strength and electrical performance of the coil, and even cause the coil to be unable to be used normally. It may also cause damage to the wire insulation layer. During the winding process, if the wire is subjected to excessive tension, the insulation layer may be damaged due to friction or stretching, which will reduce the electrical insulation performance of the coil and increase the risk of electrical faults. When the wire tension is too small, the coil may not be able to maintain a tight arrangement during the winding process, resulting in a loose coil. This loose state may affect the mechanical strength and heat dissipation performance of the coil and reduce the service life of the soft start device. Summary of the Invention

[0005] The present invention provides an intelligent winding and detection method for a three-phase integrated coil of a soft start device to solve the defect in the prior art that during the winding process, improper control of the wire tension affects the performance and life of the soft start device.

[0006] An intelligent winding and detection method for a three-phase integrated coil of a soft start device provided by the present invention includes:

[0007] S1: Determine the electrical parameters of coil winding and prepare winding materials.

[0008] S2: During the winding of the coil, adaptively control the wire tension and calculate the optimal tension within the k moment by establishing a tension mathematical model.

[0009] S3: Use a neural network model to estimate and correct the key parameters in the tension mathematical model in real time and compensate for the non-linear part.

[0010] S4: The visual inspection control center detects the quality of coil winding in real time, judges whether errors occur during the winding process, and if so, stops the winding and issues an alarm.

[0011] S5: Put the wound coil into epoxy resin glue for degassing to ensure that the insulating material evenly covers the surface of the wire and the gaps in the coil.

[0012] According to an intelligent winding and detection method for a three-phase integrated coil of a soft start device provided by the present invention, in step S2, the specific steps for adaptively controlling the wire tension are as follows:

[0013] S21: Use a tension sensor to collect the tension of the wire to obtain tension data.

[0014] S22: Preprocess the tension data to obtain preprocessed tension data.

[0015] S23: Use a model predictive control algorithm to establish a tension mathematical model for the coil winding system and calculate the optimal tension within k moments.

[0016] According to an intelligent winding and detection method for a three-phase integrated coil of a soft start device provided by the present invention, in step S22, the preprocessing includes data normalization, outlier removal, and periodicity inspection.

[0017] Perform normalization processing on the data to scale the data to a specific range.

[0018] Use the IQR method to calculate outliers. Determine the dispersion degree of the data by calculating the quartiles of the tension training data, and judge whether it is an outlier according to the set threshold. The steps for calculating outliers using the IQR method are as follows:

[0019] Sort the data set according to the numerical size.

[0020] Calculate the first quartile Q1. In the sorted data set, Q1 is the data point at the 25% position.

[0021] Calculate the third quartile Q3. In the sorted data set, Q3 is the data point at the 75% position.

[0022] Calculate IQR, IQR = Q3 - Q1.

[0023] After obtaining the IQR, judge whether the data point is an outlier according to the set threshold. If so, it is an outlier and the data is removed.

[0024] Perform periodic inspection on the data to ensure that there are no duplicate data in the data set.

[0025] A method for intelligent winding and detection of a three-phase integrated coil of a soft start device provided by the present invention. In step S23, the specific steps for establishing a tension mathematical model for the coil winding system using a model predictive control algorithm and calculating the optimal tension within k moments are as follows:

[0026] S231: In the coil winding system, determine the input variables and output variables. The input variables are the rotational speed of the motor, the torque of the motor, the wire diameter, and the number of turns, and the output variable is the tension of the wire.

[0027] S232: Based on physical principles, establish a tension mathematical model to obtain a tension change prediction formula.

[0028] S233: Predict future tension changes according to the tension change prediction formula.

[0029] S234: Set the input constraints, output constraints, and objective function of the wire tension during the test process. Use the quadratic programming algorithm to calculate the optimal output torque and rotational speed of the motor within k moments.

[0030] A method for intelligent winding and detection of a three-phase integrated coil of a soft start device provided by the present invention. In step S232, the tension change prediction formula is expressed as:

[0031]

[0032] In the formula, F is the wire tension, T is the torque, T0 is other resistance torques, d is the wire diameter, and N is the number of turns.

[0033] A method for intelligent winding and detection of a three-phase integrated coil of a soft start device provided by the present invention. In step S234, the optimal torque obtained by quadratic programming solution is expressed as:

[0034] T * (k) = U * [0]

[0035] In the formula, U * [0] is the optimal control sequence at moment k.

[0036] A method for intelligent winding and detection of a three-phase integrated coil of a soft start device provided by the present invention. In step S3, the compensation of the non-linear part that is difficult to accurately model by the neural network model for the tension mathematical model predictive control algorithm includes:

[0037] There is an error between the predicted value of the wire tension calculated by the prediction model algorithm and the actually measured wire tension.

[0038] The neural network model and the model predictive control algorithm work in parallel. The same input is respectively input into the tension change model formula of the model predictive control algorithm and the neural network model. The model predictive control algorithm outputs a prediction result according to its own linear relationship, while the neural network model outputs a corresponding non-linear compensation amount according to the non-linear mapping relationship it has learned. The linear relationship prediction result of the model predictive control algorithm and the non-linear relationship prediction result of the neural network are fused to obtain a final prediction result of linear and non-linear factors.

[0039] According to a method for intelligent winding and detection of a three-phase integrated coil of a soft start device provided by the present invention, in step S3, the specific steps for the neural network model to output a non-linear compensation amount according to the non-linear mapping relationship include:

[0040] S31: Select a multi-layer perceptron to predict the non-linear compensation amount of the tension, and initialize the connection weights and bias terms between the neurons in each layer of the neural network to obtain initialization data.

[0041] S32: The neurons in the hidden layer perform weighted summation on the initialization data and perform non-linear transformation through an activation function to obtain a predicted value of the non-linear compensation amount of the initialization data.

[0042] S33: Use the mean square error to calculate the loss function to reduce the loss of the predicted value of the non-linear compensation amount next time.

[0043] S34: Use the chain rule to calculate the gradients of the connection weights and biases of the neurons in each layer in reverse order from the output layer. According to the gradients, use the stochastic gradient descent optimization algorithm to update the parameters in the neural network.

[0044] S35: Continuously repeat the steps of weighted summation, calculating the loss function, and parameter update until the neural network model can stably output a reasonable and accurate non-linear compensation amount.

[0045] According to a method for intelligent winding and detection of a three-phase integrated coil of a soft start device provided by the present invention, in step S4, the specific steps for using the vision detection control center to monitor the winding quality in real time are as follows:

[0046] S41: The vision detection control center collects the coil image during the winding process, uses an image analysis algorithm to process the coil image, and extracts image features.

[0047] S42: According to the image features and the pre-set quality judgment parameters, judge whether the winding quality meets the requirements. If so, continue to collect and analyze the next group of images.

[0048] S43: After the winding machine control center receives the winding quality problem information fed back by the vision detection control center, it will immediately trigger a stop instruction to stop the operation of the winding machine.

[0049] S44: After the winding machine stops running, the control center of the winding machine issues an alarm.

[0050] The present invention also provides a three-phase integrated coil intelligent winding and detection system for a soft start device, including:

[0051] An electrical parameter module for determining the electrical parameters of coil winding.

[0052] An adaptive winding module for adaptively controlling the wire tension during the coil winding process and calculating the optimal tension within the k moment by establishing a tension mathematical model.

[0053] A tension data acquisition unit for using a tension sensor to collect the tension of the wire and obtain tension data.

[0054] A tension data preprocessing unit for preprocessing the tension data.

[0055] A tension calculation unit for calculating the optimal tension within the k moment using a model predictive control algorithm.

[0056] A non-linear compensation module for using a neural network model to estimate and correct the key parameters in the tension mathematical model in real time and compensate for the non-linear part.

[0057] A winding quality detection module for using a vision detection control center to detect the coil winding quality in real time, determine whether an error occurs during the winding process, and if so, stop the winding and issue an alarm.

[0058] A post-winding processing module for putting the wound coil into epoxy resin glue for defoaming to protect the wound coil.

[0059] The soft start device three-phase integrated coil intelligent winding and detection method provided by the present invention establishes a tension mathematical model for the coil winding system by using a model predictive control algorithm, calculates the optimal tension within a finite time domain, uses a neural network model to estimate and correct the key parameters in the model prediction in real time, and compensates for the non-linear part that is difficult to accurately model by the model predictive control algorithm, solving the defect that improper wire tension control during the winding process affects the performance and service life of the soft start device. The beneficial effects obtained are:

[0060] MPC is good at dealing with constraint problems and model-based dynamic prediction, while the neural network model can handle non-linear and complex unknown relationships. The combination of the two gives full play to their respective advantages, improves the adaptability of the entire tension control system to complex working conditions, and can control the tension more accurately. With the assistance of the neural network model for MPC, the prediction model can better fit the actual coil winding system, reduce the tension control deviation caused by model errors, unconsidered non-linear factors, etc., and then ensure the stability and uniformity of the tension during the entire winding process, avoiding coil winding quality problems caused by improper tension.

[0061] The neural network model can estimate and correct the key parameters in the model predictive control algorithm in real time, thus ensuring the accuracy and reliability of the tension mathematical model. The neural network model can more accurately predict the optimal tension within a finite future time domain, reducing the tension control error caused by inaccurate parameters. The tension control in the coil winding system is often affected by various non-linear factors, such as material properties, frictional resistance, etc. The neural network model can compensate for these non-linear factors, thereby improving the accuracy and stability of tension control. Combining the neural network model with the model predictive control algorithm can achieve a higher level of automatic control, reduce manual intervention, and improve production efficiency and product quality. The powerful learning and adaptation ability of the neural network model helps to enhance the adaptability and flexibility of the system, meeting the requirements of different application scenarios.

[0062] Precise tension control helps to reduce the tension fluctuation of the wire during the winding process, thus avoiding damage to the wire caused by excessive stretching or relaxation, maintaining the integrity and stability of the wire, and improving the overall performance of the soft start device. By precisely controlling the tension, the friction and wear between the wire and the winding mechanism are reduced, thereby reducing the risk of fatigue failure of the wire, extending the service life of the wire, and reducing the frequency and cost of wire replacement.

[0063] Precise tension control helps to ensure that the wire maintains a uniform tension distribution during the winding process, thereby improving the quality and consistency of winding, enhancing the performance and reliability of the soft start device, reducing the failure rate caused by winding quality problems, and reducing the failure shutdown time caused by improper tension control, thereby improving the operating efficiency of the overall system.

[0064] By precisely controlling the winding speed and tension, the insulation performance and electrical performance of the coil are guaranteed. Optimize the layout, and through optimized calculation, achieve the best layout of winding, reduce electromagnetic coupling and loss, improve the efficiency of the equipment, and improve production efficiency. Description of the Drawings

[0065] To more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0066] Figure 1 is a flowchart of the intelligent winding and detection method for the three-phase integrated coil of the soft start device provided by the embodiment of the present invention;

[0067] Figure 2 is a flowchart of the intelligent winding and detection method for the three-phase integrated coil of the soft start device provided by the embodiment of the present invention;

[0068] Figure 3 is a module of the intelligent winding and detection system for the three-phase integrated coil of the soft start device provided by the embodiment of the present invention. Detailed implementation manners

[0069] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0070] The following combines Figures 1 - 3 to describe the intelligent winding and detection method and system for the three-phase integrated coil of the soft start device of the present invention.

[0071] Figure 1 is the intelligent winding and detection method for the three-phase integrated coil of the soft start device provided by the embodiment of the present invention.

[0072] As Figures 1 - 2 shown, the intelligent winding and detection method for the three-phase integrated coil of the start device provided by the embodiment of the present invention mainly includes the following steps:

[0073] The specific steps of the intelligent winding method include:

[0074] S1: Use professional electromagnetic design software to accurately design the three-phase integrated coil, and determine parameters such as the number of turns, wire diameter, winding arrangement method, and connection method of the coil.

[0075] Prepare insulating paper, insulating sleeves, high-strength insulating materials, and binding wires before winding.

[0076] The polyester film insulation paper is used to provide insulation between the online coil layers and between the coil and the inner cylinder to prevent short circuits, and the thickness is determined according to the voltage level and insulation requirements.

[0077] Prepare insulating sleeves to wrap the lead wire part of the conductor to avoid insulation damage during connection or subsequent operations. Use epoxy resin glue for the encapsulation and fixation of the coil to ensure good insulation, heat resistance and adhesiveness.

[0078] Use fiberglass-reinforced epoxy resin material to make the inner cylinder. The size of the inner cylinder should be precisely processed according to the size requirements of the coil, and its diameter and length should ensure appropriate space and electrical performance after the coil is wound.

[0079] Binding wires are used to temporarily fix the conductors during the winding process; sandpaper is used to polish the surface of the inner cylinder to ensure its smoothness and avoid scratching the insulation layer of the conductors.

[0080] Use copper conductors to wind the coil, ensuring good electrical conductivity, mechanical strength and insulation performance of the conductors. Debug and calibrate the intelligent winding machine, and check whether all components of the winding machine are operating normally, such as the wire feeding device, wire arranging device, tension control system, counter, etc. Ensure that the accuracy and stability of the winding machine meet the winding requirements.

[0081] S2: During the process of winding the coil, adaptively control the wire tension, and calculate the optimal tension within the kth moment by establishing a tension mathematical model.

[0082] Firmly install the wire spool on the wire feeding rack, adjust the height and position of the wire feeding rack so that the wire is at the same horizontal height as the main shaft of the winding machine to facilitate the lead-out of the wire. Install the wire clamping clip between the winding machine and the wire feeding rack, clamp the magnet wire, and adjust the clamping force of the wire clamping clip so that the wire has an appropriate tension during winding. Leave an appropriate length at the starting end of the wire, generally 20 - 30 cm, depending on the specific wiring requirements and installation location. Fix the starting end on the main shaft of the winding machine in the specified manner, and use the winding method to ensure that the starting end will not loosen during the winding process.

[0083] S21: Use a tension sensor to collect the wire tension data to obtain tension training data.

[0084] S22: Preprocess the tension training data to obtain tension preprocessed data. The preprocessing includes data normalization, eliminating outliers, periodicity test and fixed change trend test.

[0085] Data normalization is to scale the data to a specific range to eliminate the differences between different data scales so that the subsequent neural network model learning can better process the data. The Z-Score normalization method is used to normalize the data so that the processed tension training data conforms to the standard normal distribution, and the formula is expressed as:

[0086]

[0087] In the formula, μ is the mean, σ is the standard deviation, and X is the tension training data.

[0088] Outlier removal is to reduce the noise and outliers in the dataset and improve the accuracy and reliability of the data. The IQR method is used to identify outliers. By calculating the quartiles of the tension training data, the degree of data dispersion is determined, and whether it is an outlier is judged according to the set threshold.

[0089] The calculation steps of IQR are as follows:

[0090] Sort the dataset according to the numerical size.

[0091] Calculate the first quartile Q1, which is the minimum value of 25% of the data points in the dataset. In the sorted dataset, Q1 is the data point at the 25% position.

[0092] Calculate the third quartile Q3, which is the minimum value of 75% of the data points in the dataset. In the sorted dataset, Q3 is the data point at the 75% position.

[0093] Calculate IQR, that is, the difference between Q3 and Q1, IQR = Q3 - Q1.

[0094] After obtaining IQR, according to the set threshold, it is judged whether the data point is an outlier. The data points lower than Q1 - 1.5×IQR or higher than Q3 + 1.5×IQR are outliers.

[0095] Periodicity test is to determine whether there is a certain repeating pattern or periodic change in the dataset. The autocorrelation function is used to perform the periodicity test on the data: by calculating the correlation between data points and themselves at different lags to judge the periodicity of the data. If the autocorrelation function shows a significant peak at a certain lag, it may indicate that the data has periodicity.

[0096] Fixed change trend test is to determine whether there is a certain fixed change trend in the dataset. The trend line fitting is used to perform the fixed change trend test on the data: methods such as linear regression and polynomial regression are used to fit the trend line of the data, and the shape and slope of the trend line are observed to judge the trend of the data.

[0097] S23: Use the model predictive control algorithm to establish a tension mathematical model for the coil winding system and calculate the optimal tension within the k-th moment.

[0098] Determine the input variables and output variables in the coil winding system. The input variables include: the rotational speed n of the motor (affecting the winding speed), the torque T of the motor (directly related to tension adjustment), wire diameter, and number of turns. These variables are adjusted by the controller, thereby affecting the tension during the coil winding process.

[0099] The output variable includes: the actual tension value F. The magnitude of the tension is directly related to the quality of the coil winding. For example, if the tension is too large, it may cause damage to the wire, and if the tension is too small, the coil winding will be loose.

[0100] Establish a model based on physical principles, and derive the tension mathematical model according to mechanical principles, kinematic principles, etc.

[0101] Consider the relationship between the rotation of the coil and the tension: Based on the relationship between tension and centripetal force and the relationship between linear velocity and angular velocity, derive Combine the above relationships with factors such as motor torque to establish an equation between variables.

[0102] The radius r of the coil will increase as the number of winding turns N increases. Its rate of change is related to the winding speed and the width b of the coil. The rate of change r of the radius r ’ The formula is expressed as:

[0103]

[0104] In the formula, b is the width of the coil.

[0105] After sorting out and discretization processing, the obtained discrete-time state space model is:

[0106]

[0107] Among them is the state vector, is the input vector, y(k) = F(k) is the output vector, and A, B, C, D are the corresponding coefficient matrices.

[0108] Based on the prediction model, in each control cycle, predict the tension change situation in the future finite time domain. Assume that during the coil winding process, at the current moment, a certain value of the coil diameter and a specific state of the motor rotational speed are detected. Input these data into the prediction model, and the model can, according to its internal mathematical relationships and parameters, calculate how the tension will change in the future period as the winding continues.

[0109] Let the dependent variable be the wire tension, and the independent variables be the rotational speed, torque, wire diameter, and number of turns of the motor to obtain the tension change prediction formula. The formula is expressed as:

[0110]

[0111] Wherein, F is the wire tension, T is the torque, T0 is other resistance torques, d is the wire diameter, and N is the number of turns.

[0112] There are various constraints in the actual system. For example, the torque provided by the motor has upper and lower limit ranges, which limits the amplitude of tension adjustment; the unwinding and rewinding speeds cannot exceed the safe operating speed range of the equipment; and the increase in the coil diameter is also within a certain physically realizable range, etc.

[0113] Input constraint conditions:

[0114] u min ≤ T ≤ u max

[0115] Wherein, i = 0, 1, …, Nc.

[0116] Output constraint conditions:

[0117] F min ≤ F ≤ F max

[0118] Wherein, i = 0, 1, …, Np.

[0119] Clarify the goal of tension control to keep the tension stable within a set optimal range. For example, for the winding of a certain high-precision electronic coil, it is required that the tension error be controlled within ±1N. At the same time, it may also be expected that the rate of tension change should also be maintained within a certain reasonable range to avoid the impact of sudden tension changes on the winding quality.

[0120] The formula expression of the objective function is:

[0121]

[0122] Wherein, q is the output weight, r is the control increment weight, y(k) is the tension, r(k) is the tension reference value, u(k) is the control motor torque, N p is the prediction time domain length, N c is the control time domain length, and N c ≤ N p , △u(k+i|k) = u(k+i|k) - u(k+i-1|k) represents the control increment, that is, the change amount of the motor torque relative to the previous moment at each moment.

[0123] Combined with the set objective function and constraint conditions, the quadratic programming algorithm is used to calculate the optimal control sequence within the current moment to a finite future time domain. During the winding process of the three-phase integrated coil, according to the predicted tension change, the tension objective function, and the constraint conditions, the optimal sequence of control quantities such as the torque and speed that the motor should output at each moment in the next few seconds is calculated through the optimization algorithm, so that the tension can be as close as possible to and maintained within the preset optimal range.

[0124] The formula for calculating the optimal torque through quadratic programming is expressed as:

[0125] T * (k) = U * [0]

[0126] In the formula, U is the optimal control sequence at time k.

[0127] Directly extract the first element from the obtained optimal control sequence vector, which is the optimal torque value at the current moment k. This value is the result of the optimization algorithm that minimizes the objective function after comprehensively considering the current state of the system, the desired tension, and various constraint conditions.

[0128] For the optimal torque at k + i, extract the corresponding element from the optimal control sequence U* in the same way. Its expression is:

[0129] T * (k + i|k) = U * [i]

[0130] Take the element with index i in the U* vector. This element corresponds to the optimal torque value at time k + i, which reflects the optimal torque that should be adopted at each future moment pre-calculated through the optimization algorithm based on the information at the current moment k. By using this to control the motor, the tension can be as close as possible to and maintained within the set optimal range during the future operation of the system, while meeting various constraint conditions related to torque.

[0131] S3: The neural network model estimates and corrects the key parameters in the tension mathematical model in real time and compensates for the nonlinear part.

[0132] Compare the actual tension value with the predicted tension value output by the prediction model, and calculate the deviation between the two. If the deviation exceeds the allowable range, it indicates that the prediction model may not match the actual system, which may be due to unconsidered interference factors during the operation of the equipment or certain modeling errors in the model itself, etc. Use this deviation information to correct the prediction model so that the subsequent prediction results can be closer to the actual tension change situation, thereby improving the control accuracy.

[0133] The neural network model has a powerful non - linear mapping ability. Using the observable input - output data of the system as training samples, it mines the hidden relationships therein, and then estimates these key parameters. Through the internal neuron connections and activation functions, the neural network model performs complex non - linear operations in the hidden layer. Finally, it outputs the estimated values related to the key parameters in the output layer and continuously corrects them according to the actual situation to ensure the accuracy of the estimation. In each control cycle of MPC, there will be a deviation between the predicted value of the system output calculated based on the model prediction algorithm and the actually measured system output. This deviation information is fed back into the neural network model to further adjust parameters such as the weights and thresholds of the neural network model, thereby realizing the correction of the estimated values of the key parameters.

[0134] The neural network model works in parallel with the linear prediction model of MPC. The same input is sent into the linear prediction model of MPC and the neural network model respectively. The linear prediction model of MPC outputs the prediction result according to its own linear relationship, while the neural network model outputs the corresponding non - linear compensation amount according to the non - linear mapping relationship it has learned. These two results are appropriately fused to obtain a final prediction result that comprehensively considers linear and non - linear factors for subsequent optimization and control decisions.

[0135] S31: Select a multi - layer perceptron to predict the non - linear compensation amount of tension based on the current input. The multi - layer perceptron includes an input layer, several hidden layers, and an output layer. For example, the number of nodes in the input layer contains 5 features such as speed and material properties, so there are 5 nodes in the input layer. Set 2 - 3 hidden layers, with 10 and 8 neurons respectively in each layer, and 1 node in the output layer. Initialize the connection weights and bias terms between neurons in each layer of the neural network. These initial parameters determine the initial computing ability and expression ability of the network, and will be gradually optimized through training later.

[0136] S32: Send the initialized data into the input layer of the neural network. The data is sequentially passed from the input layer to the hidden layer. In each hidden layer, the neurons perform weighted summation on the initialized data and perform non - linear transformation through the activation function. For the ReLU activation function, for a certain neuron j in the hidden layer, the formula for the weighted summation of the input x is:

[0137]

[0138] In the formula, ω ij is the weight connecting the previous - layer node i to this neuron j, x i is the input value of the previous layer, and b j is the bias of this neuron.

[0139] Then, through the activation function a j= max(0, z j ) to obtain the output a j , and this output will serve as the input for the next layer of neurons. Calculate layer by layer in this way to achieve feature extraction and non - linear transformation of the input data.

[0140] After multiple calculations in the hidden layer, the data is finally passed to the output layer. The neurons in the output layer also perform calculations such as similar weighted summation. Finally, the predicted value corresponding to the non - linear compensation amount of the input data is output. In the task of predicting the non - linear compensation amount of tension, the nodes in the output layer will output a specific value or a vector, and this value or vector is the preliminary estimate of the non - linear compensation amount given by the neural network based on the current input.

[0141] S33: Select an appropriate loss function according to the nature of the output task to measure the difference between the non - linear compensation amount of the network output and the actual "true" compensation amount. For the specific compensation amount value of predicting tension, the calculation formula of the mean squared error (MSE) loss function is:

[0142]

[0143] In the formula, n is the number of samples, y i is the actual value, is the network predicted value.

[0144] Substitute the non - linear compensation amount of the network output and the corresponding true compensation amount in the training data into the selected loss function to calculate the loss value after the current forward propagation. For example, in a batch of training samples, calculate the loss value corresponding to this batch using the mean squared error formula based on the network output and the actual value. This loss value will be used as the basis for subsequent adjustment of the network parameters, hoping that the next output of the network can make the loss smaller.

[0145] S34: Based on the calculated loss value, use the chain rule to calculate the gradients of the connection weights and biases of the neurons in each layer in reverse order from the output layer. For example, for a certain weight ω in the output layer, its gradient value is obtained through a series of derivative operations This gradient represents the degree and direction of the influence of this parameter on the loss value. Similarly, calculate the gradients of other relevant parameters in each layer. These gradient information are the key to guiding how to adjust the subsequent parameters.

[0146] According to the calculated gradients, use the stochastic gradient descent optimization algorithm to update the parameters in the network. The parameter update formula of the stochastic gradient descent optimization algorithm is:

[0147]

[0148] In the formula, η is the learning rate, which determines the step size of parameter update.

[0149] By continuously adjusting parameters such as weights and biases according to the results of backpropagation of gradients in this way, the network will gradually learn a more accurate mapping relationship between the input and the non-linear compensation amount, making the non-linear compensation amount output subsequently closer and closer to the actual situation.

[0150] S35: Repeat the above steps of weighted summation, calculating the loss function, and parameter update for multiple rounds. Each round will use different batch data in the training set. As the number of iterations increases, the network's learning of the non-linear mapping relationship of the input data will become deeper and deeper, and the accuracy of the output non-linear compensation amount will also gradually improve.

[0151] During the training process, use the validation set data to verify the model regularly, and calculate the loss value or other evaluation metrics on the validation set. Observe the changes in the metrics of the validation set. If it is found that the loss value of the validation set starts to increase as the number of training rounds increases, corresponding measures can be taken, such as stopping training in advance, increasing the regularization term to adjust the model, to ensure that the model has good generalization ability and can stably output a reasonable and accurate non-linear compensation amount.

[0152] S36: After the above sufficient training and verification, in the actual application scenario, for the newly input data related to tension, the neural network can, according to the learned mapping relationship, through the forward propagation step, output the corresponding non-linear compensation amount at the output layer, providing corresponding values for subsequent operations such as fusing with the results of the model predictive control algorithm, thereby improving the overall prediction effect.

[0153] S4: The visual detection control center detects the coil in real time, determines whether an error occurs during the winding process, and if so, stops the winding and issues an alarm.

[0154] Regularly check the appearance of the coil. The visual detection system will detect the winding quality in real time, such as the number of turns, wire diameter, insulation performance, etc. Once problems are found, such as broken wire ends, damaged insulation layers, incorrect number of turns, whether the wires are arranged neatly, and whether there are intersections, the winding machine will automatically stop and issue an alarm signal. The operator will perform corresponding processing according to the alarm information, such as repairing the broken ends, replacing the insulation material, and adjusting the winding parameters.

[0155] S41: After the winding machine starts winding the coil, the vision inspection system continuously acquires the coil images during the winding process at a set frequency. For each frame of the acquired images, it is processed using the built-in image analysis algorithm. First, the images are preprocessed to remove noise, enhance contrast, etc., so as to more clearly extract features subsequently. Key features related to the winding quality are extracted. For example, the contour of the wire is identified through the edge detection algorithm, and then it is judged whether the wire arrangement is neat and whether it exceeds the preset spacing range; the morphological algorithm is used to detect abnormal conditions such as wire breakage and knotting; different-phase windings are distinguished through image segmentation technology, and it is checked whether the connection parts are correct, etc.

[0156] S42: Based on the extracted features and the preset quality judgment parameters, the vision inspection system judges the winding quality in real time. If the analysis result shows that the current winding state meets the quality requirements, the system continues to normally acquire and analyze the next set of images; once a problem is detected during winding, such as the wire arrangement spacing is too large exceeding the allowable error, obvious wire breakage is found, or the three-phase winding connection is incorrect, etc., the vision inspection system immediately feeds back the inspection results including details such as the problem type and occurrence location to the winding machine control system through the communication link.

[0157] S43: After receiving the winding quality problem information fed back by the vision inspection system, the winding machine control system immediately triggers a stop instruction, and by controlling relevant components such as the power supply or driver of the motor, makes the winding machine stop running quickly, avoiding more quality defects caused by continuous winding, and minimizing raw material waste and subsequent rework costs as much as possible.

[0158] S44: When the alarm signal is triggered, the winding machine control system activates the alarm device to emit the corresponding alarm signal. The alarm signal is in the form of an audible and visual alarm. For example, the warning light is lit and a continuous beeping sound is emitted simultaneously to attract the attention of the operator, facilitating the operator to promptly know that a problem has occurred during winding and come to check and handle it. Parameters such as the intensity and frequency of the alarm signal are reasonably set according to factors such as the noise level of the actual working environment and the distance between the operator and the winding machine to ensure that the abnormal information can be effectively conveyed.

[0159] S5: The wound coil is placed in epoxy resin glue for degassing to ensure that the insulating material evenly covers the surface of the wire and the gaps in the coil.

[0160] S51: Use binding wires to bind the wound coil as a whole. The binding spacing is 5 - 10 cm to firmly fix the coil and prevent wire displacement during subsequent operations. Cut off the excess binding wires during the winding process to make the coil appearance neat. Check the coil appearance to ensure that there are no loose or protruding wires. For untidy wires, gently adjust them with tools to make them neatly arranged. Withdraw the winding mold and take out the wound three - phase integrated coil, avoiding collision and scratching the coil surface.

[0161] S52: Process the lead copper bars into appropriate shapes, remove the oxide layer and impurities on the surface to ensure good electrical conductivity. Clean and weld the wire ends of each phase to ensure firm and reliable connection between the wire ends. Use soldering equipment to weld the three lead copper bars on each coil to the corresponding branch wire ends respectively, ensuring firm welding without false soldering. After welding, check the welding points, measure the resistance with a multimeter to ensure reliable connection. Provide insulation protection for the welding points and lead copper bars. Use insulating sleeves or insulating tapes to wrap the connection parts and exposed parts of the lead copper bars to prevent short - circuit or leakage. Arrange the lead copper bars according to the design requirements to ensure reasonable positions for subsequent connection to the external circuit, and maintain a certain spacing between the lead copper bars to avoid mutual interference or short - circuit.

[0162] S53: Stir the epoxy resin glue evenly according to a certain ratio, and then put it into a vacuum device for degassing treatment to remove the air bubbles in the glue, ensuring that the insulating material evenly covers the wire surface and the gaps in the coil to avoid insulation weak points. Place the coil in the injection mold, and inject the degassed epoxy resin glue into the mold through the injection hole to make the glue fully fill the gaps in the coil, ensuring that the coil is completely surrounded by the glue. The injection process should be slow and uniform to prevent the generation of air bubbles. Place the injected coil in a suitable environment and cure it according to the curing requirements of the epoxy resin glue. Generally, it requires a certain temperature and time, such as curing at 60 - 80 °C for 2 - 4 hours to ensure that the epoxy resin glue is fully cured to form a solid whole. Fix the fixing plate to the coil through a bolt structure, ensuring that the bolt tightening torque meets the requirements, so that the front - end epoxy board and the rear - end epoxy board clamp the inner cylinder in the middle, and the rear - end epoxy sealing board is closely attached to the rear - end epoxy board for fixation. During the installation process, pay attention to insulating the fixing plate, such as using insulating gaskets to avoid conduction with the coil or the inner cylinder.

[0163] As Figure 3 shown, the present invention also provides a three - phase integrated coil intelligent winding and detection system for a soft - start device, including:

[0164] An electrical parameter module for determining the electrical parameters of coil winding.

[0165] An adaptive winding module, which is used to adaptively control the wire tension during the winding process of the coil, and calculates the optimal tension within the k moment by establishing a tension mathematical model.

[0166] A tension data acquisition unit, which is used to collect the tension of the wire using a tension sensor to obtain tension data.

[0167] A tension data preprocessing unit, which is used to preprocess the tension data.

[0168] A tension calculation unit, which is used to calculate the optimal tension within the k moment using a model predictive control algorithm.

[0169] A non-linear compensation module, which is used to use a neural network model to estimate and correct the key parameters in the tension mathematical model in real time, and compensate for the non-linear part.

[0170] A winding quality detection module, which is used to use a vision detection control center to detect the winding quality of the coil in real time, judge whether an error occurs during the winding process, and if so, stop winding and issue an alarm.

[0171] A post-winding processing module, which is used to put the wound coil into epoxy resin glue for degassing to protect the wound coil.

[0172] The intelligent winding and detection method for the three-phase integrated coil of the soft start device provided by the present invention, by using a model predictive control algorithm to establish a tension mathematical model for the coil winding system, calculates the optimal tension within a finite time domain, and a neural network model estimates and corrects the key parameters in the model prediction in real time, and compensates for the non-linear part that is difficult to accurately model by the model predictive control algorithm, solves the defect that the improper control of the wire tension during the winding process affects the performance and service life of the soft start device. The beneficial effects obtained are:

[0173] MPC is good at dealing with constraint problems and model-based dynamic prediction, while the neural network model can handle non-linear and complex unknown relationships. The combination of the two gives full play to their respective advantages, improves the adaptability of the entire tension control system to complex working conditions, and can control the tension more accurately. With the assistance of the neural network model for MPC, the prediction model can better fit the actual coil winding system, reduce the tension control deviation caused by model errors, unconsidered non-linear factors, etc., and then ensure the stability and uniformity of the tension during the entire winding process, and avoid the coil winding quality problems caused by improper tension.

[0174] By accurately controlling the winding speed and tension, the insulation performance and electrical performance of the coil are ensured. Optimize the layout, and through optimization calculations, achieve the best layout of the winding, reduce electromagnetic coupling and losses, and improve the efficiency of the equipment. High degree of automation, reduce manual intervention, and improve production efficiency.

[0175] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A three-phase integrated coil intelligent winding and detection method for a soft start device, characterized in that: include: S1: Determine the electrical parameters of the coil winding and prepare the winding materials; S2: During the coil winding process, the wire tension is adaptively controlled, and the optimal tension within time k is calculated by establishing a tension mathematical model; S21: Using a tension sensor to collect the tension of the wire to obtain tension data; S22: preprocessing the tension data to obtain tension preprocessing data; S23: Use the model predictive control algorithm to establish a tension mathematical model for the coil winding system and calculate the optimal tension within time k; S231: In the coil winding system, determining input variables and output variables; the input variables are the speed of the motor, the torque of the motor, the wire diameter and the number of turns, and the output variable is the tension of the wire; S232: Establishing a tension mathematical model based on physical principles to obtain a tension change prediction formula; S233: Calculating the future tension according to the tension change prediction formula to predict the future tension change; The tension change prediction formula is expressed as: In the formula, F is the wire tension, T is the torque, T0 is other resistance torque, d is the wire diameter, and N is the number of turns; S234: setting input constraints, output constraints and objective function of the wire tension during the test; using the quadratic programming algorithm to calculate the optimal output torque and speed of the motor within k moments; The optimal torque obtained by solving the quadratic programming is expressed as: T * (k)=U * [0] Where U * [0] is the optimal control sequence at time k; The first element is directly extracted from the optimal control sequence vector obtained by solving, which is the optimal torque value at the current time k; For the optimal torque of k+i, the corresponding element is extracted from the optimal control sequence U*, and its expression is: T * (k+i|k)=U * [i] Take the element with index i in the U* vector, which corresponds to the optimal torque value at time k+i, and use the optimization algorithm to pre-calculate the optimal torque to be used at each future moment; S3: using a neural network model to estimate and correct key parameters in the tension mathematical model in real time, and to compensate for nonlinear parts; S4: The visual inspection control center detects the coil winding quality in real time and determines whether there is an error in the winding process. If so, the winding process is stopped and an alarm is issued; S5: Place the wound coil into epoxy resin glue for degassing, ensuring that the insulating material is evenly covered on the surface of the wire and in the gap between the coil.

2. The three-phase integrated coil intelligent winding and detection method for a soft start device according to claim 1 is characterized in that: In step S22, the preprocessing includes data normalization, outlier removal, and periodic line inspection; Normalize the data and scale it to a specific range; The IQR method is used to calculate wild points. The degree of data dispersion is determined by calculating the quartiles of the tension training data, and whether it is a wild point is determined based on the set threshold. The steps for calculating wild points using the IQR method are as follows: Sort the data set by numerical size; Calculate the first quartile Q1. In the sorted data set, Q1 is the data point at the 25% position. Calculate the third quartile Q3. In the sorted data set, Q3 is the data point at the 75% position. Calculate IQR, IQR = Q3 - Q1; After obtaining the IQR, the set threshold is used to determine whether the data point is an outlier. If so, the data point is removed. Perform periodic checks on the data to ensure that there is no duplicate data in the data set.

3. The three-phase integrated coil intelligent winding and detection method for a soft starter device according to claim 1 is characterized in that: In step S3, the neural network model compensates for the nonlinear part that is difficult to accurately model by the tension mathematical model predictive control algorithm, including: There is an error between the wire tension prediction value calculated by the prediction model algorithm and the wire tension actually measured; The neural network model and the model predictive control algorithm work in parallel, and the same input is input into the tension change model formula of the model predictive control algorithm and the neural network model respectively. The model predictive control algorithm outputs the prediction result according to its own linear relationship, and the neural network model outputs the corresponding nonlinear compensation amount according to the nonlinear mapping relationship it has learned. The linear relationship prediction result of the model predictive control algorithm is fused with the nonlinear relationship prediction result of the neural network to obtain a final prediction result of linear and nonlinear factors.

4. The three-phase integrated coil intelligent winding and detection method for a soft starter device according to claim 1 is characterized in that: In step S3, the specific steps of outputting the nonlinear compensation amount of the nonlinear mapping relationship of the neural network model include: S31: Select a multilayer perceptron to predict the nonlinear compensation amount of tension, and initialize the connection weights and bias items between neurons in each layer of the neural network to obtain initialization data; S32: The neurons in the hidden layer perform weighted summation on the initialization data and perform nonlinear transformation through an activation function to obtain a predicted value of a nonlinear compensation amount of the initialization data; S33: using the mean square error to calculate the loss function, and reducing the loss of the predicted value of the next nonlinear compensation amount; S34: using the chain rule to calculate the gradients of the connection weights and biases of the neurons in each layer in reverse order from the output layer, and using a stochastic gradient descent optimization algorithm to update the parameters in the neural network according to the gradients; S35: Repeat the weighted summation, loss function calculation and parameter update steps until the neural network model can stably output a reasonable and accurate nonlinear compensation amount.

5. The three-phase integrated coil intelligent winding and detection method for a soft starter device according to claim 1 is characterized in that: In step S4, the specific steps of using the visual inspection control center to monitor the winding quality in real time are as follows: S41: The visual inspection control center collects the coil image during the winding process, processes the coil image using an image analysis algorithm, and extracts image features; S42: judging that the winding quality meets the requirements according to the image features and the preset quality judgment parameters, and if so, continuing to collect and analyze the next set of images; S43: After receiving the winding quality problem information fed back by the visual inspection control center, the winding machine control center will immediately trigger a stop command to stop the winding machine; S44: After the winding machine stops running, the winding machine control center sends out an alarm.

6. A three-phase integrated coil intelligent winding and detection system for a soft starter device, which adopts the three-phase integrated coil intelligent winding and detection method for a soft starter device as claimed in any one of claims 1 to 5, characterized in that: The intelligent winding system comprises: Electrical parameter module, used to determine the electrical parameters of coil winding; The adaptive winding module is used to adaptively control the wire tension during the coil winding process and calculate the optimal tension within k moments by establishing a tension mathematical model; A tension data acquisition unit, used to acquire the tension of the conductor using a tension sensor to obtain tension data; A tension data preprocessing unit, used for preprocessing the tension data; A tension calculation unit, used to calculate the optimal tension within k moments using a model predictive control algorithm; A nonlinear compensation module, used to use a neural network model to estimate and correct key parameters in the tension mathematical model in real time and to compensate for the nonlinear part; The winding quality detection module is used to detect the coil winding quality in real time using the visual detection control center to determine whether there is an error in the winding process. If so, the winding process is stopped and an alarm is issued; The post-winding processing module is used to place the wound coil into epoxy resin glue for degassing and protect the wound coil.

Images