Oral removable denture support deformation reverse compensation method based on PSO-BP (Particle Swarm Optimization-Back Propagation) network

Abstract

The invention discloses a PSO-BP network-based anti-deformation compensation method for an oral removable denture support, and belongs to the field of selective laser melting metal printing. The method comprises the steps of obtaining deformation data of the removable denture support, building and training a PSO-BP network, building an anti-compensation model of the removable denture support, and printing the anti-compensation model. Simulating an SLM printing process through commercial finite element additive manufacturing, and exporting a deformation result; the deformation result serves as PSO-BP network input, and the original model serves as network output to train the network; the test network uses an original model as input, and an output target is a reverse compensation model; network output is packaged into an STL format through three-dimensional scanning and reverse engineering software, and a packaged result is sliced and printed; the neural network is mainly used for learning the deformation rule, so that the method has a good application prospect in selective laser melting metal printing.

Description

Technical Field

[0001] The present invention belongs to the field of selective laser melting metal printing; specifically, it relates to a deformation anti-compensation method for removable dental prosthesis brackets based on a PSO-BP network. Background Art

[0002] Selective laser melting technology (SLM) has been widely applied to fields such as aviation, aerospace, and medicine due to its short printing time, high printing efficiency, and ability to print multi-material parts with complex structures. In stomatology, removable partial dentures are a very common restoration method, especially for most middle-aged and elderly people. The development of digital oral scanning in collaboration with intelligent manufacturing technologies such as SLM has made the production method of removable denture brackets simpler and faster, and has shown good applicability in clinical practice.

[0003] SLM is an additive manufacturing technology that achieves the repair or manufacture of metal parts by layer-by-layer scanning and printing of the sliced model of the removable denture bracket, and through powder melting and deposition. Compared with the traditional production method of denture brackets, it can achieve rapid prototyping and improve material utilization rate. However, during the printing and forming process, factors such as the rapid increase and cooling of temperature, and the addition of support structures, which cannot be changed, cause the parts to warp and deform during the forming process. The deformation of the partial denture bracket will cause problems such as chewing difficulties, speech disorders, nausea, and discomfort in the mandibular joint for the patient, and will have a greater impact on the subsequent treatment of the doctor. Therefore, reducing the deformation of the removable partial denture bracket during the SLM printing process is an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a deformation anti-compensation method for removable dental prosthesis brackets based on a PSO-BP network, which can reduce the deformation of the oral bracket and improve the printing accuracy.

[0005] To achieve the above objectives, the present invention adopts the following technical solutions:

[0006] A deformation anti-compensation method for removable dental prosthesis brackets based on a PSO-BP network, comprising the following steps:

[0007] S1: Obtaining the deformation data of the removable denture bracket, using commercial finite element additive manufacturing to simulate the denture printing process, and exporting the STL file format of the deformed model;

[0008] S2: Building and training a PSO-BP network, determining that the numbers of input, output, and hidden layers are 4, 1, and 9 respectively; initializing the PSO algorithm parameters;

[0009] S3: Converting the STL file of the deformed model into the X-T file format, and using it as the input of the network, and using the X-T file of the original model as the output of the network to train the network;

[0010] S4: After the network training is completed, take the original data X-T file of the removable denture bracket as the input, and the output target is the data of the anti-compensation model of the removable denture bracket, and export the X-T file data of the anti-compensation model of the removable denture bracket;

[0011] S5: Slice and print the anti-compensation model, encapsulate the anti-compensation model data into the STL format through 3D scanning and reverse engineering software, and perform slice printing.

[0012] In the above steps, step S2 specifically includes the following steps:

[0013] S2-1: Build and train the PSO-BP network, initialize the parameters of the BP network, and initialize the velocity, position, individual / group optimal values of the particles;

[0014] S2-2: Update the particle velocity and position, determine the historical optimum of the individual / global, and judge whether the global fitness is less than the set accuracy; if so, jump to S2-3; if not, judge whether the number of iterations is greater than the specified maximum number of iterations; if so, output the global optimal particle position; if not, perform another iteration until it is greater than the maximum number of iterations;

[0015] S2-3: Take the deformed model after simulation as the input and the original model as the output to train the network;

[0016] S2-4: Calculate the output and error of each layer, and judge whether the error requirement is met; if so, the training ends; if not, calculate the reverse error, update the weight threshold, and calculate the error again until the error requirement is met;

[0017] S2-5: Input the original model data into the trained BP network and output the reverse compensation model.

[0018] Beneficial effects: The present invention provides a method for anti-compensation of deformation of an oral removable denture bracket based on a PSO-BP network. Compared with the existing technology, it has the following advantages:

[0019] High precision: The BP network anti-compensation method based on the particle swarm algorithm can achieve high precision and can provide a more accurate solution in complex deformation anti-compensation problems;

[0020] Strong applicability: This method can be applied to various deformation anti-compensation problems of oral removable denture brackets, and has a wide range of applications;

[0021] High calculation efficiency: This method uses the particle swarm algorithm for optimization, can obtain the optimal solution in a short time, and has higher calculation efficiency;

[0022] High reliability: The BP network anti-compensation method based on the particle swarm optimization algorithm has been verified through multiple experiments, has high reliability, and has good experimental results;

[0023] Combining the above advantages, the method for anti-compensating the deformation of the removable dental prosthesis bracket provided by the present invention can provide better technical support for the field of stomatology. Brief Description of the Drawings

[0024] Figure 1 It is the flowchart of the anti-compensation of the removable partial denture in the embodiment of the present invention;

[0025] Figure 2 It is the flowchart of the commercial finite element additive manufacturing simulation in the embodiment of the present invention;

[0026] Figure 3 It is the flowchart of the PSO-BP network anti-compensation in the embodiment of the present invention. Detailed Description of the Preferred Embodiments

[0031] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments:

[0032] As Figures 1 - 3 shown, a method for anti-compensating the deformation of a removable dental prosthesis bracket based on a PSO-BP network includes the following steps:

[0033] S1: Obtaining the deformation data of the removable dental prosthesis bracket, using commercial finite element additive manufacturing to simulate the dental prosthesis printing process, and exporting the STL file format of the deformed model;

[0034] S2: Building and training a PSO-BP network. The BP neural network mainly consists of neurons in the input layer, hidden layer, and output layer. Determine that the numbers of neurons in the input, output, and hidden layers are 4, 1, and 9 respectively. Neurons in each layer are connected by weights and thresholds. Initialize the BP network parameters, the velocity, position, individual / group optimal values of the particles. Through research, it is determined that there are 4 neurons in the input layer, and only one neuron in the output layer. The 4 neurons in the input layer are laser power, scanning speed, scanning spacing, and placement angle. There is no limit to the number of layers in the hidden layer, but the number of its neurons can be limited by the following formula:

[0035]

[0036] where l is the number of neurons in the hidden layer, N is the number of training samples, m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is a constant between 5 and 10;

[0037] After conducting multiple experiments in combination with the above empirical formula, it is found that when the number of neurons in the hidden layer is selected as 9, the convergence speed is relatively fast and the error is relatively small during the training of the neural network. Therefore, the topological structure of the finally established BP model is 4×9×1;

[0038] S3: Update the particle velocity and position, determine the historical optimum of the individual / global, and judge whether the global fitness is less than the set accuracy; if so, jump to S4; if not, judge whether the number of iterations is greater than the specified maximum number of iterations; if so, output the global optimum particle position; if not, perform another iteration until it is greater than the maximum number of iterations. Optimize the BP network parameters through the PSO algorithm. PSO is an optimization algorithm used to solve many different types of problems; in PSO, each particle represents a possible solution, and the entire particle swarm represents all possible solutions to the problem. The following are the formulas used in the PSO algorithm optimization:

[0039] (1) Velocity update formula:

[0040] v(i) = w × v i +c 1 ×rand()×(pbest i -x i )+c 2 ×rand()×(gbest - x i )

[0041] Where: v(i) is the velocity of particle i; w is the inertia weight; vi is the current velocity of particle i; c1 and c2 are acceleration constants; rand() is a random function that generates a random number between [0,1]; pbest i is the personal best position of particle i; gbest is the global best position; x i is the current position of particle i;

[0042] (2) Position update formula:

[0043] x(i) = xi + vi

[0044] Where: x i is the current position of particle i; v i is the current velocity of particle i;

[0045] Generally speaking, the velocity update formula determines the direction of particle movement towards the global best and individual best positions, while the position update formula directly determines the new position of the particle. These two formulas combined form the main process of the particle swarm optimization algorithm;

[0046] S5: Use the simulated deformed model as the input and the original model as the output to train the network; calculate the output and error of each layer, and determine whether the error requirement is met; if so, the training ends; if not, calculate the reverse error, update the weight threshold, and calculate the error again until the error requirement is met;

[0047] Through the forward propagation method, information enters the network from the input layer, passes through the calculations of each layer in turn, and obtains the final output layer result. Multiply the value of each layer by the corresponding weight + bias variable (activation function);

[0048] From the input layer to the hidden layer:

[0049] From the hidden layer to the output layer:

[0050] Among them, v and w are the weights from the input layer to the hidden layer and from the hidden layer to the output layer respectively; x is the neuron of the input layer; b is the neuron of the hidden layer;

[0051] Calculate the error between the output layer and the expected value through the backpropagation method to adjust the network parameters, so as to make the error smaller;

[0052] The error calculation formula is as follows:

[0053] Weight reverse update: wi = (1)Eyy k w = △w, + w

[0054] Among them, y is the value of the output layer; T is the expected value; i is the serial number of the neuron in the input layer; l is the learning rate, which can adjust the update pace. A suitable learning rate can make the objective function converge to the local minimum within a suitable time, and is generally selected as 0.01 - 0.8;

[0055] S6: Input the original model data into the trained BP network and output the reverse compensation model;

[0056] S7: Convert the STL file of the deformed model into the X-T file format and use it as the input of the network, and use the X-T file of the original model as the output of the network to train the network;

[0057] S8: After the network training is completed, use the X-T file of the original data of the removable denture bracket as the input, and the output target is the data of the reverse compensation model of the removable denture bracket, and export the X-T file data of the reverse compensation model of the removable denture bracket;

[0058] S9: Slice and print the reverse compensation model, encapsulate the reverse compensation model data into the STL format through 3D scanning and reverse engineering software, and perform slice printing.

[0059] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention cannot be limited thereby. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for anti-compensation of removable denture bracket deformation based on PSO-BP network, characterized in that: The steps include: S1: Obtain deformation data of removable denture bracket, simulate denture printing process, and export the deformed model; S2: Build and train the PSO-BP network and initialize the PSO algorithm parameters; S3: construct a removable denture framework anti-compensation model, take the removable denture framework original data as input, and output the removable denture framework anti-compensation model data; S4: Slice and print the anti-compensation model.

2. The method for anti-compensation of deformation of removable denture bracket based on PSO-BP network according to claim 1, characterized in that: Commercial finite element additive manufacturing was used to simulate the denture printing process in S1.

3. The method for anti-compensation of deformation of removable denture bracket based on PSO-BP network according to claim 1, characterized in that: The number of input, output and hidden layers of the PSO-BP network described in S2 is 4, 1 and 9 respectively.

4. The method for anti-compensation of deformation of removable denture bracket based on PSO-BP network according to claim 1 or 3, characterized in that: S2 specifically includes the following steps: S2-1: Build and train the PSO-BP network, initialize the BP network parameters, initialize the particle speed, position, and individual / group optimal values; S2-2: Update particle speed and position, determine individual / global historical optimality, and judge whether the global fitness is less than the set accuracy; if so, jump to S2-3; if not, judge whether the number of iterations is greater than the specified maximum number of iterations; if so, output the global optimal particle position; if not, perform another iteration until it is greater than the maximum number of iterations; S2-3: Use the simulated deformed model as input and the original model as output to train the network; calculate the output and error of each layer to determine whether the error requirement is met; if so, the training ends; if not, calculate the reverse error, update the weight threshold, and calculate the error again until the error requirement is met; S2-4: Input the original model data into the trained BP network and output the reverse compensation model.

5. The method for anti-compensation of removable denture bracket deformation based on PSO-BP network according to claim 4, characterized in that: The BP neural network is composed of neurons in the input layer, hidden layer and output layer, and the neurons in each layer are connected through weights and thresholds.

6. The method for anti-compensation of removable denture bracket deformation based on PSO-BP network according to claim 5, characterized in that: The input layer has 4 neurons, the output layer has 1 neuron, and the number of hidden layers is unlimited. The number of neurons can be limited by the following formula: Among them, l is the number of neurons in the hidden layer, N is the number of training samples, m is the number of neurons in the input layer, n is the number of neurons in the output layer, and a is a constant between 5 and 10.

7. The method for anti-compensation of removable denture bracket deformation based on PSO-BP network according to claim 4, characterized in that: The formula for the update speed in S2-2 is: v (i) =w×v i +c1×rand()×(pbest i -x i )+c2×rand()×(gbest-x i ) Where: v (i) is the velocity of particle i; w is the inertia weight; v i is the current speed of particle i; c1 and c2 are acceleration constants; rand() is a random function that generates a random number between [0, 1]; pbest i is the personal best position of particle i; gbest is the global best position; x i is the current position of particle i.

8. The method for anti-compensation of removable denture bracket deformation based on PSO-BP network according to claim 4, characterized in that: The formula for updating the position in S2-2 is: x (i) =x i +v i Where: x i is the current position of particle i; v i is the current velocity of particle i.

9. The method for anti-compensation of deformation of removable denture bracket based on PSO-BP network according to claim 4, characterized in that: In S2-3, the output and error of each layer are calculated. The information is transmitted from the input layer into the network through the forward propagation method. After each layer is calculated in turn, the final output layer result is obtained. From input layer to hidden layer: From hidden layer to output layer: Among them, v and w are the weights from the input layer to the hidden layer, and from the hidden layer to the output layer respectively; x is the neuron of the input layer; and b is the neuron of the hidden layer.

10. The method for anti-compensation of deformation of removable denture bracket based on PSO-BP network according to claim 4, characterized in that: In S2-3, the reverse error is calculated by back propagation to calculate the error between the output layer and the expected value to adjust the network parameters, so that the error becomes smaller. The error calculation formula is as follows: Weight reverse update: Δw i =(l)E yk In i =Δw i +in i Among them, y is the value of the output layer, T is the expected value, i is the sequence number of the input layer neuron, and l is the learning rate.

Images