Laser marking method for ophthalmic surgical instrument

Abstract

The invention provides a laser marking method for an ophthalmic surgical instrument, which comprises the following steps: (a) data acquisition: carrying out a preliminary laser marking experiment on the surface of a target material for the ophthalmic surgical instrument, and acquiring laser marking parameters and effect data corresponding to the marking parameters to obtain a marking data set; (b) data preprocessing: preprocessing the marking data set; the preprocessing comprises the steps of removing noise, filling missing values and mapping the marking data to the same scale by using Max-Min normalization; (c) neural network model establishment: setting parameters of a BP neural network to establish a BP neural network model; (d) training and optimizing: training the data obtained in the step (b) for multiple times by using the BP neural network model; and (e) parameter optimization and verification. The optimal parameter setting is determined after multiple times of training, the marking effect is ensured, meanwhile, the surface integrity of the instrument is protected to the maximum degree, and the marking precision is effectively improved.

Description

Technical Field

[0001] The invention belongs to the technical field of medical instruments and relates to a laser marking method, in particular to a laser marking method for ophthalmic surgical instruments. Background Art

[0002] Ophthalmic surgical instruments refer to medical devices used for ophthalmic surgery, mainly including surgical scissors, forceps, tweezers, hooks and eyelid openers. As an important category of medical devices, the quality and precision of ophthalmic surgical instruments are crucial to the effectiveness of ophthalmic surgery. Product quality traceability refers to the process of ensuring product quality by recording and tracking the production, circulation and sales of products in detail (usually using a one-to-one digital ID card as a carrier). If the quality traceability of ophthalmic surgical instruments can be achieved, the name, time and place of the person responsible can be clarified, responsibilities can be clearly defined, and investigations and punishments can be based on evidence, which greatly enhances the sense of responsibility of employees.

[0003] Laser marking is a technology that uses a high-energy laser beam to mark the surface of a material. Its principle is to form a permanent mark on the surface of the material through the thermal effect or photochemical reaction of the laser beam. Compared with traditional marking technologies such as inkjet printing and embossing, laser marking has the advantages of non-contact, pollution-free, high precision, and long-lasting marking. Therefore, it is widely used in high-precision and high-requirement fields (such as medical device manufacturing). In order to ensure the performance and traceability of medical devices, laser marking is usually performed on the surface of medical devices during the manufacturing process to identify the brand, model, batch and other key information. However, since medical devices are mostly made of high-strength alloys or special medical materials, these materials have significant differences in absorption and reaction to laser energy. If the laser parameters are not set properly, the surface of the medical device may be damaged, affecting its mechanical properties and biocompatibility. Therefore, how to optimize the laser marking parameters to ensure clear marking without affecting the use of medical devices has become an urgent problem to be solved.

[0004] The Chinese invention patent with application number 202010718686.3 discloses a medical device laser marking system and method, which includes the following steps: obtaining the logo pattern to be marked through the control terminal; setting the parameters of the laser beam, and controlling the laser to emit a laser beam with corresponding parameters; controlling the laser beam to mark a black mark on the surface of the medical device in a focused or defocused manner according to the logo pattern to be marked, so as to obtain a medical device with a black mark. This application mainly realizes rapid heat treatment of the surface material of the medical device by controlling the energy density of the laser beam, so that the surface of the medical machine is permanently blackened to obtain a medical device with a striking black mark. This is different from the purpose of forming traceable information on the surface of ophthalmic surgical instruments. In addition, there is a lack of precise control of the control terminal, which makes it difficult to meet the high standards of ophthalmic surgical instruments. Summary of the invention

[0005] In view of the above-mentioned defects, the present invention provides a laser marking method for ophthalmic surgical instruments, which can automatically find the optimal laser marking parameters, thereby improving the accuracy and consistency of marking.

[0006] In order to achieve the above technical objectives, the present invention provides a laser marking method for ophthalmic surgical instruments, which is characterized by comprising the following steps: (a) Data collection: A preliminary laser marking experiment was conducted on the surface of the target material used for ophthalmic surgical instruments, and the laser marking parameters and the effect data under the corresponding marking parameters were collected to obtain the marking data set; (b) Data preprocessing: preprocessing the labeled data set; the preprocessing includes removing noise, filling missing values, and mapping the labeled data to the same scale using Max-Min normalization; (c) Establishing a neural network model: Setting parameters of a BP neural network to establish a BP neural network model; the parameters of the BP neural network include the number of neurons in the hidden layer, the number of hidden layers, the probability that a neuron stops transmitting data, a learning rate, hyperparameters of an Adam optimizer, initial weights and biases, and an activation function; (d) Training and optimization: the data obtained in step (b) is trained multiple times using the BP neural network model; (e) Parameter optimization and verification: The BP neural network model is used to predict the marking effects under different parameter combinations, and the marking accuracy of each parameter combination is calculated and recorded; when the error between the predicted value of the marking accuracy and the actual value of the marking accuracy is the smallest, the corresponding marking parameters are used to perform laser marking on the ophthalmic surgical instruments.

[0007] Optimally, in step (a), the marking parameters include at least laser power, marking speed and focal diameter, and the effect data includes at least a predicted value of marking accuracy and an actual value of marking accuracy.

[0008] Furthermore, in step (c), a BN layer is added to the BP neural network model to ensure the stability of weight and bias parameters in each iteration.

[0009] Furthermore, in step (d), the network weights and biases are adjusted during training to optimize the predictive ability of the model; The RMSE loss function is used to measure the error between the predicted value of the marking accuracy and the actual value of the marking accuracy of the BP neural network model, and the parameters are adjusted through the Adam optimizer to accelerate convergence.

[0010] Specifically, in step (e), the calculation of the marking accuracy is performed according to the following steps: A weighted sum algorithm and an activation function are used to convert high-dimensional data into a predicted value of marking accuracy of the BP neural network model; the RMSE loss function is used to calculate the error between the actual value of marking accuracy and the predicted value of marking accuracy. When the loss function value has not reached the global minimum, back propagation is performed, and the weights and bias values ​​of the neurons are adjusted using the Adam optimizer, and then the forward propagation algorithm is performed until the RMSE loss function value stops decreasing; when the marking accuracy still cannot meet the requirements after back propagation, the hyperparameters of the Adam optimizer are iteratively adjusted until the error between the predicted value of marking accuracy and the actual value of marking accuracy is minimized; the determination coefficient is used to measure the prediction accuracy, and the optimal marking machine parameters and the weights and bias values ​​of the BP neural network model are saved.

[0011] Furthermore, in step (c), the BP neural network comprises an input layer, two hidden layers, two BN layers and an output layer. The calculation formula from the input layer to the hidden layer is: ; In the formula, is the weighted input of the hidden layer, is the weighted output of the hidden layer, f is the set activation function, is the weight matrix, is the bias vector; The calculation formula from the input layer to the output layer is: ; In the formula, is the weighted input to the output layer, is the weight between the last hidden layer and the output layer, is the bias between the last hidden layer and the output layer; is the activation output of the activation layer, and g is the activation function of the output layer.

[0012] Specifically, in step (e), the RMSE loss function is calculated according to the following formula: , In the formula, y is the actual value of marking accuracy, is the predicted value of marking accuracy.

[0013] Specifically, in step (e), the gradient of the output layer is calculated by back propagation using the following formula:

[0014] Specifically, in step (e), the gradient descent method is used to adjust the weights and bias values ​​of the neurons in the BP neural network model according to the following formula to minimize the loss function value: ; ; In the formula, is the adjusted weight of the BP neural network model, is the weight of the BP neural network model before adjustment, is the bias value after adjustment of the BP neural network model, It is the bias value before the BP neural network model is adjusted; is a hyperparameter of the model, which is the learning rate; is the partial derivative of the loss function with respect to the weight, is the partial derivative of the loss function with respect to the bias.

[0015] Optimally, in step (b), the calculation formula of the Max-Min normalization is: ; Where x is the entire labeled dataset, is a value in the data set, min(x) is the minimum value in the data set, max(x) is the maximum value in the data set, is the normalized value.

[0016] The laser marking method for ophthalmic surgical instruments of the present invention fully utilizes the nonlinear learning ability of the BP neural network and constructs a complex relationship between laser marking parameters and marking accuracy, thereby avoiding calculation deviations caused by different parameter scales, determining the optimal parameter settings after multiple trainings, reducing the sample scrap rate caused by human experience deviations, ensuring the marking effect, and protecting the surface integrity of the instrument to the greatest extent, thereby effectively improving the marking accuracy, consistency and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the laser marking method of the ophthalmic surgical instrument of the present invention; Figure 2 This is a BP neural network operation diagram in the laser marking method for ophthalmic surgical instruments of the present invention; Figure 3 This is a marking diagram of a surface of a metal sheet (stainless steel) using the laser marking method of the ophthalmic surgical instrument of the present invention; Figure 4 This is a marking diagram of the other surface of a metal sheet (stainless steel) using the laser marking method of the ophthalmic surgical instrument of the present invention; Figure 5 This is a correlation diagram between current and marking clarity of the present invention; Figure 6 This is a correlation diagram between the number of processing times and the marking clarity of the present invention; Figure 7 This is a correlation diagram between the marking speed and the marking clarity of the present invention; Figure 8 This is a correlation diagram between the Q pulse width and the marking clarity of the present invention; Fig. 9 This is a correlation diagram between frequency and marking clarity of the present invention; Fig.10 It is a loss function variation diagram of the present invention; Fig.11 This is a comparison chart of the marking results of stainless steel and titanium alloy of the present invention; Fig.12 This is a marking diagram of a sample of an ophthalmic medical device (tweezers) of the present invention. DETAILED DESCRIPTION

[0018] The present invention is further described in detail below in conjunction with specific examples, but the present invention is not limited to the following examples. The implementation conditions used in the examples can be further adjusted according to different requirements of specific use, and the implementation conditions not specified are conventional conditions in the industry.

[0019] like Figure 1 The laser marking method for ophthalmic surgical instruments shown comprises the following steps: (a) Data collection: Conduct preliminary laser marking experiments on the surface of target materials for ophthalmic surgical instruments, collect laser marking parameters and effect data under corresponding marking parameters to obtain marking data sets; marking parameters at least include key parameters such as laser power, marking speed and focal diameter (specifically, current, number of processing times, marking speed, Q pulse width and frequency, etc.); effect data (i.e., marking quality data or marking accuracy data, such as marking clarity, etc.) at least include predicted values ​​of marking accuracy and actual values ​​of marking accuracy; (b) Data preprocessing: Preprocess the labeled data set; preprocessing includes removing noise, filling missing values, and mapping the labeled data to the same scale using Max-Min normalization; (c) Establishment of neural network model: Set the parameters of BP neural network to establish BP neural network model (referred to as model); the parameters of BP neural network include the number of neurons in hidden layer, the number of hidden layers, the probability that neurons stop transmitting data, learning rate, hyperparameters of Adam optimizer, initial weight (i.e. initial weight of neurons) and bias (i.e. bias of neurons) and activation function (sigmoid function). Adding BN layer to BP neural network model ensures the stability of weight and bias parameters in each iteration, thus avoiding the problem of gradient explosion.

[0020] (d) Training and optimization: The data obtained in step (b) (i.e., the pre-processed marking data) is trained multiple times using the BP neural network model. During training, the network weights and biases are adjusted to optimize the model's prediction ability. The RMSE loss function is used to measure the error between the marking accuracy prediction value and the actual marking accuracy value of the BP neural network model. The parameters are adjusted through the Adam optimizer to accelerate convergence and avoid falling into a local optimal solution.

[0021] (e) Parameter optimization and verification: The BP neural network model is used to predict the marking effect under different parameter combinations, and the marking accuracy of each parameter combination is calculated and recorded; when the error between the predicted value of marking accuracy and the actual value of marking accuracy is the smallest, the corresponding marking parameters are used to perform laser marking on ophthalmic surgical instruments. If the prediction error of the model does not reach the expected accuracy, the hyperparameters of the model (such as learning rate, first-order moment estimation, second-order moment estimation, weight decay value, etc.) are further adjusted and the model is retrained until the prediction result meets the accuracy requirements. The accuracy and generalization ability of the model are verified using the test set, and the model parameters are analyzed to ensure that the optimized parameter combination can stably achieve high-quality laser marking in practical applications.

[0022] Specifically, the calculation of marking accuracy is carried out according to the following steps: using a weighted sum algorithm and an activation function to convert high-dimensional data into a marking accuracy prediction value of the BP neural network model; using the RMSE loss function to calculate the error between the actual value of marking accuracy and the predicted value of marking accuracy: when the loss function value has not reached the global minimum, back propagation is performed, and the weights and bias values ​​of the neurons are adjusted using the Adam optimizer, and then the forward propagation algorithm is performed until the RMSE loss function value stops decreasing; when the marking accuracy still cannot meet the requirements after back propagation, the hyperparameters of the Adam optimizer are iteratively adjusted until the error between the predicted value of marking accuracy and the actual value of marking accuracy is minimized; the determination coefficient is used to measure the prediction accuracy, and the optimal marking machine parameters and the weights and bias values ​​of the BP neural network model are saved.

[0023] In step (c) of this embodiment, the BP neural network includes an input layer, two hidden layers, two BN layers and an output layer. The calculation formula from the input layer to the hidden layer is: ; In the formula, is the weighted input of the hidden layer, is the weighted output of the hidden layer, f is the set activation function, is the weight matrix, is the bias vector; The calculation formula from the input layer to the output layer is: ; In the formula, is the weighted input to the output layer, is the weight between the last hidden layer and the output layer, is the bias between the last hidden layer and the output layer; is the activation output of the activation layer, and g is the activation function of the output layer.

[0024] In step (e), the RMSE loss function is calculated according to the following formula: , In the formula, y is the actual value of marking accuracy, Prediction of marking accuracy.

[0025] In step (e), the gradient of the output layer is calculated by back propagation using the following formula:

[0026] In step (e), the gradient descent method is used to adjust the weights and bias values ​​of the neurons in the BP neural network model according to the following formula to minimize the loss function value: ; ; In the formula, is the adjusted weight of the BP neural network model, is the weight of the BP neural network model before adjustment, is the bias value after adjustment of the BP neural network model, It is the bias value before the BP neural network model is adjusted; is a hyperparameter of the model, which is the learning rate; is the partial derivative of the loss function with respect to the weight, is the partial derivative of the loss function with respect to the bias.

[0027] In step (b), the calculation formula of the Max-Min normalization is: ; Where x is the entire labeled dataset, is a value in the data set, min(x) is the minimum value in the data set, max(x) is the maximum value in the data set, is the normalized value.

[0028] The number of neurons in the two hidden layers is set to 100. The data set is divided into a training set and a test set. The training set is used for BP neural network model training (the number of training times is 2000 times), and the test set is used to verify the learning ability and prediction accuracy of the model.

[0029] The activation function is the Sigmoid function: ; In the formula, x is the output value after weighted operation of the previous layer, It is the output value after the nonlinear transformation of Sigmoid function.

[0030] The formula for the BN layer is as follows: ; In the formula, is the i-th sample of the input, m is the number of samples in the current mini-batch, is the mean of the mini-batch, is the variance of the mini-batch, is the standardized sample, is a learnable scaling factor, is the learnable offset, Small constant to prevent division by zero errors.

[0031] The optimizer is the Adam optimizer: ; In the formula, t is the number of iterations, is the momentum of the tth iteration, is the momentum of the t-1th iteration, , are the hyperparameter exponential decay rates (used to ensure smooth gradient changes between two adjacent iterations), S t is the exponentially weighted moving average variable of the tth iteration, S t-1 is the exponentially weighted moving average variable of the t-1th iteration, is the parameter gradient value obtained in the previous iteration, , Respectively and Bias correction to prevent the weight from oscillating violently when the gradient approaches the global minimum. is the parameter that needs to be gradient-decreased for the tth iteration, is the parameter that needs to be gradient-decreased in the t-1th iteration, is a constant added to maintain numerical stability, is the model's hyperparameter learning rate (set by the model user).

[0032] The RMSE loss function is used to measure the error between the predicted value and the theoretical value. , where n is the number of samples, is the true value of the i-th sample, is the predicted value of the ith sample.

[0033] The laser marking method of the above-mentioned ophthalmic surgical instruments was used for experiments, as follows: Each set of data contains 5 parameters and an output value. The parameters include frequency (10-99KHz), current intensity (1~20A), number of processing times (1~15 times), marking speed (10~200mm / s), Q pulse width; and the material type (stainless steel (such as Figure 3 and Figure 4The output value is the clarity of the manually verified marking result, and the result is between [1~6]. First, 15 experiments were conducted on the frequency, current, number of processing times, Q pulse width, and marking speed parameters to calculate the control variables. The experimental results Figures 5 to 9 shown.

[0034] The fixed materials are stainless steel and titanium alloy. The orthogonal experimental design method is used to generate 80 sets of random data for the remaining five parameters using a computer. The data are shown in Table 1.

[0035] Table 1 Random data table when the fixing material is stainless steel and titanium alloy

[0036] Table 1

[0037] The above data were input into the laser marking machine, and 80 experiments were conducted on the stainless steel plate and the titanium alloy plate respectively. The experimental results of the stainless steel plate are as follows: Figure 2 and Figure 3 As shown, the results for titanium alloy plates are similar.

[0038] The marking results are divided into six levels, as shown in Table 2, and the evaluation level is used as the output parameter of the data set to construct the data set required by the BP neural network.

[0039] Table 2 Marking result grade table

[0040] The BP neural network is used to train the data set. After 2000 trainings, the model reaches the global optimal solution, such as Fig.10 Then the marking results of stainless steel and titanium alloy with the same input parameters were compared. Fig.11 As shown; the experimental results show that, on stainless steel plates and titanium alloy plates with the same thickness, marking on titanium alloy steel plates generally has higher clarity.

[0041] After training is completed, obtain the parameter matrix of the model's weight W and bias b and save it. Set the user to obtain a specific output (such as a clarity of 4), and use the gradient descent optimization algorithm to adjust the input parameters so that the network's output is as close to the target value as possible. Convert this problem into an optimization problem, namely: , where the input parameter x is the variable to be obtained, W and b are the weight value and output value of the neural network model respectively, and f is the functional relationship between W, b and x.

[0042] Definition of loss function: The loss function Loss is the square of the difference between the target output and the current output. By minimizing this loss function, the input parameter combination closest to the target output can be found; Based on the trained model, the laser marking machine was set to the global optimal parameters, and the marking of ophthalmic medical device samples was verified to prove the reliability of the method. The results are as follows Fig.12 shown.

[0043] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A laser marking method for ophthalmic surgical instruments, characterized in that: The following steps are involved: (a) Data collection: A preliminary laser marking experiment was conducted on the surface of the target material used for ophthalmic surgical instruments, and the laser marking parameters and the effect data under the corresponding marking parameters were collected to obtain the marking data set; (b) Data preprocessing: preprocessing the labeled data set; the preprocessing includes removing noise, filling missing values, and mapping the labeled data to the same scale using Max-Min normalization; (c) Establishing a neural network model: Setting parameters of a BP neural network to establish a BP neural network model; the parameters of the BP neural network include the number of neurons in the hidden layer, the number of hidden layers, the probability that a neuron stops transmitting data, a learning rate, hyperparameters of an Adam optimizer, initial weights and biases, and an activation function; (d) Training and optimization: the data obtained in step (b) is trained multiple times using the BP neural network model; (e) Parameter optimization and verification: The BP neural network model is used to predict the marking effects under different parameter combinations, and the marking accuracy of each parameter combination is calculated and recorded; when the error between the predicted value of the marking accuracy and the actual value of the marking accuracy is the smallest, the corresponding marking parameters are used to perform laser marking on the ophthalmic surgical instruments.

2. The laser marking method for ophthalmic surgical instruments according to claim 1, characterized in that: In step (a), the marking parameters include at least laser power, marking speed and focal point diameter, and the effect data include at least a predicted value of marking accuracy and an actual value of marking accuracy.

3. The laser marking method for ophthalmic surgical instruments according to claim 2, characterized in that: In step (c), a BN layer is also added to the BP neural network model to ensure the stability of weight and bias parameters in each iteration.

4. The laser marking method for ophthalmic surgical instruments according to claim 3, characterized in that: In step (d), the network weights and biases are adjusted during training to optimize the model’s predictive power; The RMSE loss function is used to measure the error between the predicted value of the marking accuracy and the actual value of the marking accuracy of the BP neural network model, and the parameters are adjusted through the Adam optimizer to accelerate convergence.

5. The laser marking method for ophthalmic surgical instruments according to claim 4, characterized in that: In step (e), the calculation of the marking accuracy is performed according to the following steps: A weighted sum algorithm and an activation function are used to convert high-dimensional data into a predicted value of marking accuracy of the BP neural network model; the RMSE loss function is used to calculate the error between the actual value of marking accuracy and the predicted value of marking accuracy. When the loss function value has not reached the global minimum, back propagation is performed, and the weights and bias values ​​of the neurons are adjusted using the Adam optimizer, and then the forward propagation algorithm is performed until the RMSE loss function value stops decreasing; when the marking accuracy still cannot meet the requirements after back propagation, the hyperparameters of the Adam optimizer are iteratively adjusted until the error between the predicted value of marking accuracy and the actual value of marking accuracy is minimized; the determination coefficient is used to measure the prediction accuracy, and the optimal marking machine parameters and the weights and bias values ​​of the BP neural network model are saved.

6. The laser marking method for ophthalmic surgical instruments according to claim 3, characterized in that: In step (c), the BP neural network comprises an input layer, two hidden layers, two BN layers and an output layer. The calculation formula from the input layer to the hidden layer is: ; In the formula, is the weighted input of the hidden layer, is the weighted output of the hidden layer, f is the set activation function, is the weight matrix, is the bias vector; The calculation formula from the input layer to the output layer is: ; In the formula, is the weighted input to the output layer, is the weight between the last hidden layer and the output layer, is the bias between the last hidden layer and the output layer; is the activation output of the activation layer, and g is the activation function of the output layer.

7. The laser marking method for ophthalmic surgical instruments according to claim 4 or 5, characterized in that: In step (e), the RMSE loss function is calculated according to the following formula: ， In the formula, y is the actual value of marking accuracy, is the predicted value of marking accuracy.

8. The laser marking method for ophthalmic surgical instruments according to claim 7, characterized in that: In step (e), the gradient of the output layer is calculated by back propagation using the following formula: ， In the formula, is the loss function over the activation output The gradient of is the derivative of the activation function, is the element-wise multiplication symbol, is the error term of the output layer.

9. The laser marking method for ophthalmic surgical instruments according to claim 8, characterized in that: In step (e), the gradient descent method is used to adjust the weights and bias values ​​of the neurons in the BP neural network model according to the following formula to minimize the loss function value: ； ； In the formula, is the adjusted weight of the BP neural network model, is the weight of the BP neural network model before adjustment, is the bias value after adjustment of the BP neural network model, It is the bias value before the BP neural network model is adjusted; is a hyperparameter of the model, which is the learning rate; is the partial derivative of the loss function with respect to the weight, is the partial derivative of the loss function with respect to the bias.

10. The laser marking method for ophthalmic surgical instruments according to claim 1, characterized in that: In step (b), the calculation formula of the Max-Min normalization is: ； Where x is the entire labeled dataset, is a value in the data set, min(x) is the minimum value in the data set, max(x) is the maximum value in the data set, is the normalized value.

Images