技术领域technical field
本发明涉及神经网络技术领域，尤其涉及一种基于神经网络的太阳能电池硅片切割参数预测方法。The invention relates to the technical field of neural networks, in particular to a method for predicting cutting parameters of silicon wafers of solar cells based on neural networks.
背景技术Background technique
近些年来国家逐渐开始重视发展绿色产业，太阳能电池则是一项利用太阳能发电的环保绿色产业。在太阳能电池生产中，将硅锭切割成晶片(切片)的加工成本占太阳能电池组件生产总成本的28％。近几十年来，由于金刚石线切割机(DWS)具有生产能力高、材料消耗低、厚度测量精准和切割的硅片表面粗糙度(Ra)低的特点，因此使用DWS切割硅锭成为目前太阳能电池硅片加工领域的常规方法。DWS在加工时，切割时间较长且需要大量价格昂贵的切割液，而未优化的切割参数会导致切割时间过长和切割的硅片表面不够平整。切割时间过长导致太阳能硅片切割液大量使用，硅片表面不够平整直接影响研磨操作的加工时间，导致能源、材料过度消耗和加工效率低下。因此，必须精准确定DWS的切割参数，以优化材料和能源消耗，最小化Ra的值以控制总工艺加工时间。In recent years, the country has gradually begun to attach importance to the development of green industries, and solar cells are an environmentally friendly and green industry that uses solar power to generate electricity. In solar cell production, the processing cost of cutting silicon ingots into wafers (slicing) accounts for 28% of the total cost of solar cell module production. In recent decades, due to the high production capacity, low material consumption, precise thickness measurement and low surface roughness (Ra) of the cut silicon wafers, the use of DWS to cut silicon ingots has become the current trend for solar cells. A conventional method in the field of silicon wafer processing. During DWS processing, the cutting time is long and a large amount of expensive cutting fluid is required, while unoptimized cutting parameters will lead to long cutting time and uneven surface of the cut silicon wafer. Excessive cutting time leads to the use of a large amount of cutting fluid for solar silicon wafers. The surface of the silicon wafer is not smooth enough to directly affect the processing time of the grinding operation, resulting in excessive consumption of energy and materials and low processing efficiency. Therefore, it is necessary to accurately determine the cutting parameters of DWS to optimize material and energy consumption, and minimize the value of Ra to control the total process processing time.
由于参数较多且每个参数可选的范围较大，而控制变量的方法需要测量大量数据，极大的浪费材料和时间，并且各组数据之间的关系较为复杂，不易查找出最优切割参数。Due to the large number of parameters and the wide range of options for each parameter, the method of controlling variables needs to measure a large amount of data, which greatly wastes materials and time, and the relationship between each group of data is relatively complicated, so it is difficult to find the optimal cutting parameter.
发明内容Contents of the invention
本发明所要解决的技术问题是如何提供一种能够利用少量的实验数据进行模拟仿真，精确有效的预测出各个切割参数与表面粗糙度之间对应关系的方法。The technical problem to be solved by the present invention is how to provide a method that can accurately and effectively predict the corresponding relationship between each cutting parameter and surface roughness by using a small amount of experimental data for simulation.
为解决上述技术问题，本发明所采取的技术方案是：一种基于神经网络的太阳能电池硅片切割参数预测方法，其特征在于，实现步骤如下：In order to solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for predicting cutting parameters of solar cell silicon wafers based on neural network, characterized in that, the implementation steps are as follows:
采集样本数据：Collect sample data:
选择不同的切割参数，利用金刚石线锯切割机对直径为100～120mm的硅锭进行切割操作，测量硅片表面粗糙度Ra，获得样本数据；Select different cutting parameters, use a diamond wire saw cutting machine to cut silicon ingots with a diameter of 100-120mm, measure the surface roughness Ra of silicon wafers, and obtain sample data;
其中，所述切割参数包括：金刚线的线速度、Z轴切割进给速度和太阳能硅片切割液纯度；Wherein, the cutting parameters include: the linear speed of the diamond wire, the Z-axis cutting feed speed and the purity of the solar wafer cutting fluid;
其中所述样本数据分为三部分，70％的样本数据用作训练神经网络，15％的样本数据用作验证神经网络，15％的样本数据用作测试神经网络；Wherein the sample data is divided into three parts, 70% of the sample data is used for training the neural network, 15% of the sample data is used for verifying the neural network, and 15% of the sample data is used for testing the neural network;
训练神经网络模型：Train a neural network model:
隐藏层的输入函数为：The input function of the hidden layer is:
隐藏层的输出函数为：hk＝f(net_hidden)；The output function of the hidden layer is: hk = f(net_hidden);
输出层的输入函数为：The input function of the output layer is:
输出层的输出函数为：Oz＝f(net_output)＝RaThe output function of the output layer is: Oz = f(net_output) = Ra
其中，J为输入层神经元的个数，为Cj,k是输入神经元和隐含神经元之间的权重，ij是归一化后的输入参数包括金刚线的线速度、Z轴切割进给速度和太阳能硅片切割液纯度，θk是隐藏节点的偏置，K为隐藏层神经元的个数，Dk,z是隐藏神经元和输出神经元之间的权重，hk是隐藏节点的输出值，是输出节点的偏置，z为输出层神经元个数，f是tansig传递函数Among them, J is the number of neurons in the input layer, is Cj, k is the weight between the input neuron and the hidden neuron, and ij is the input parameter after normalization, including the linear velocity of the diamond wire, the Z axis Cutting feed speed and purity of solar wafer cutting fluid, θk is the bias of hidden nodes, K is the number of neurons in the hidden layer, Dk, z is the weight between hidden neurons and output neurons, hk is the output value of the hidden node, is the bias of the output node, z is the number of neurons in the output layer, f is the tansig transfer function
将采集样本数据中收集到的训练数据归一化后用于训练神经网络，利用验证数据调整优化网络的结构，计算期望输出与实际输出的误差，若所有误差都达到标准，则完成神经网络训练，否则进行反向传播计算隐层误差并更新权值，经过反复学习训练直至误差达到限定标准；Normalize the training data collected in the sample data and use it to train the neural network, use the verification data to adjust and optimize the structure of the network, and calculate the error between the expected output and the actual output. If all the errors meet the standard, the neural network training is completed. , otherwise carry out backpropagation to calculate the hidden layer error and update the weight value, after repeated learning and training until the error reaches the limit standard;
其中神经网络具有三层结构，输入层节点个数为3个，隐藏层节点个数为 10个，输出层节点个数为1个；The neural network has a three-layer structure, the number of input layer nodes is 3, the number of hidden layer nodes is 10, and the number of output layer nodes is 1;
预测切割参数，对比误差并找到最优参数配置：Predict the cutting parameters, compare the errors and find the optimal parameter configuration:
利用训练完成的神经网络预测模型进行预测，计算样本测量数据与神经网络仿真结果之间的相对误差，并对其他未测量参数进行预测，通过比对表面粗糙度与选择的切割参数的关系，获得表面粗糙度对应切割参数的最优解。Use the trained neural network prediction model to make predictions, calculate the relative error between the sample measurement data and the neural network simulation results, and predict other unmeasured parameters. By comparing the relationship between the surface roughness and the selected cutting parameters, it is obtained The surface roughness corresponds to the optimal solution of the cutting parameters.
进一步的技术方案在于，所述采集样本数据中切割参数选择范围为：A further technical solution is that the selection range of cutting parameters in the collected sample data is:
金刚线的线速度2.5～5.0m/s，Z轴切割进给速度0.5～2mm/min，金刚线切割液纯度为100％～30％。The linear speed of the diamond wire is 2.5-5.0m/s, the Z-axis cutting feed speed is 0.5-2mm/min, and the purity of the diamond wire cutting fluid is 100%-30%.
进一步的技术方案在于，所述采集样本数据中每片硅片的表面粗糙度Ra，是通过测量硅片表面九处不同区域后取均值获取，表面粗糙度的计算公式如下：A further technical solution is that the surface roughness Ra of each silicon chip in the collected sample data is obtained by measuring nine different areas on the surface of the silicon chip and taking the average value. The calculation formula of the surface roughness is as follows:
其中，函数Y(x)为硅片剖面表面粗糙度曲线，L为采样长度，横坐标x为采样点位置，单位为毫米，纵坐标Y为表面粗糙度，单位为微米。Among them, the function Y(x) is the surface roughness curve of the silicon wafer section, L is the sampling length, the abscissa x is the position of the sampling point in millimeters, and the ordinate Y is the surface roughness in microns.
进一步的技术方案在于，所述训练神经网络模型中采用了将切割参数归一化的方法，所用归一化公式为：A further technical solution is that a method of normalizing the cutting parameters is adopted in the training neural network model, and the normalization formula used is:
其中，ij是归一化后的切割参数数据，dmax是切割参数样本数据的最大值， dmin是切割参数样本数据的最小值，dj是第j个切割参数样本数据。Among them, ij is the normalized cutting parameter data, dmax is the maximum value of the cutting parameter sample data, dmin is the minimum value of the cutting parameter sample data, and dj is the jth cutting parameter sample data.
进一步的技术方案在于，所述训练神经网络模型中神经网络选择的是前馈反传方法，使用MATLAB的ANN工具箱软件包仿真时的参数选择为：Further technical scheme is, what neural network selects in the described training neural network model is the feed-forward backpropagation method, and the parameter selection when using the ANN toolbox software package simulation of MATLAB is:
训练函数选用Trainlm，传递函数选用Tansig，学习函数选用Learngdm，性能评价函数选用均方误差，最大训练次数选择为1000，ANN的学习率为0.01，最初的权重和偏置由MATLAB自动导出。The training function is Trainlm, the transfer function is Tansig, the learning function is Learngdm, the performance evaluation function is mean square error, the maximum number of training times is 1000, the learning rate of ANN is 0.01, and the initial weight and bias are automatically derived by MATLAB.
进一步的技术方案在于，所述预测切割参数中相对误差的计算公式为：A further technical solution is that the formula for calculating the relative error in the predicted cutting parameters is:
其中，实验结果为采集样本数据中所测实验数据，预测结果为神经网络的预测数据。Wherein, the experimental result is the experimental data measured in the collected sample data, and the predicted result is the predicted data of the neural network.
采用上述技术方案所产生的有益效果在于：本发明所述方法构建的人工神经网络，能够根据不同切割参数条件下获得的少量实验数据训练神经网络，在训练完成的人工神经网络中确定并模拟新的切割参数条件，以预测新切割参数条件下的Ra的值，避免了为获取足够多的实验数据而进行大量实验操作，节省了人力和物力消耗。根据预测结果能够得到对应特定Ra值的最佳切削参数，以使Ra、切割时间、材料和能量消耗最小化，降低了生产成本，达到了最终的提效节能的目的。The beneficial effect produced by adopting the above-mentioned technical scheme is that the artificial neural network constructed by the method of the present invention can train the neural network according to a small amount of experimental data obtained under different cutting parameter conditions, and determine and simulate new cutting parameters in the trained artificial neural network. Cutting parameter conditions, to predict the value of Ra under the new cutting parameter conditions, avoiding a large number of experimental operations to obtain enough experimental data, saving manpower and material resources. According to the prediction results, the optimal cutting parameters corresponding to a specific Ra value can be obtained, so as to minimize Ra, cutting time, material and energy consumption, reduce production costs, and achieve the ultimate goal of improving efficiency and saving energy.
附图说明Description of drawings
下面结合附图和具体实施方式对本发明作进一步详细的说明。The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.
图1本发明实施例所述预测方法的流程图；Fig. 1 is a flow chart of the prediction method described in the embodiment of the present invention;
图2本发明实施例所述预测方法中神经网络结构图；Fig. 2 neural network structural diagram in the prediction method described in the embodiment of the present invention;
图3本发明实施例所述预测方法中神经网络流程图；Fig. 3 neural network flowchart in the prediction method described in the embodiment of the present invention;
图4本发明实施例所述预测方法中切割参数-Ra的预测曲线图。Fig. 4 is a prediction curve diagram of the cutting parameter -Ra in the prediction method according to the embodiment of the present invention.
具体实施方式Detailed ways
下面结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明的一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。The technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.
在下面的描述中阐述了很多具体细节以便于充分理解本发明，但是本发明还可以采用其他不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本发明内涵的情况下做类似推广，因此本发明不受下面公开的具体实施例的限制。In the following description, a lot of specific details are set forth in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described here, and those skilled in the art can do it without departing from the meaning of the present invention. By analogy, the present invention is therefore not limited to the specific examples disclosed below.
本发明的目的是提供一种基于神经网络的硅片切割参数预测方法，对目前生产工艺中常用的金刚石线锯切割机进行参数预测，具体实施步骤如下：The purpose of the present invention is to provide a method for predicting parameters of silicon wafer cutting based on neural network, and to predict the parameters of the diamond wire saw cutting machine commonly used in the current production process. The specific implementation steps are as follows:
步骤1，采集样本数据。Step 1, collect sample data.
选择不同的切割参数，利用金刚石线锯切割机对直径为100mm的硅锭进行28次切割操作，测量硅片表面粗糙度Ra，共获取28组样本数据；其中切割参数包括：线速度、Z轴速度和切割液纯度。切割参数选择范围为：线速度2.5～ 5.0m/s，Z轴速度0.5～2mm/min，切割液纯度为100％和30％。为了提高硅片表面粗糙度Ra的精准度，所采用的方法是测量同一片硅片表面的九个不同区域粗糙度后取均值得到。表面粗糙度的计算公式如下：Choose different cutting parameters, use the diamond wire saw cutting machine to cut silicon ingots with a diameter of 100mm 28 times, measure the surface roughness Ra of the silicon wafers, and obtain 28 sets of sample data in total; the cutting parameters include: line speed, Z axis Speed and cutting fluid purity. The selection range of cutting parameters is: line speed 2.5-5.0m/s, Z-axis speed 0.5-2mm/min, cutting fluid purity 100% and 30%. In order to improve the accuracy of the surface roughness Ra of the silicon wafer, the method adopted is to measure the roughness of nine different areas on the surface of the same silicon wafer and then take the average value. The calculation formula of surface roughness is as follows:
其中，函数Y(x)为硅片剖面表面粗糙度曲线，L为采样长度，横坐标x为采样点位置，单位为毫米，纵坐标Y为表面粗糙度，单位为微米。Among them, the function Y(x) is the surface roughness curve of the silicon wafer section, L is the sampling length, the abscissa x is the position of the sampling point in millimeters, and the ordinate Y is the surface roughness in microns.
表1所示的则是不同切割参数条件下测量的28组实验样本数据。Table 1 shows the data of 28 groups of experimental samples measured under different cutting parameter conditions.
将样本数据分为三部分，其中70％的样本数据用作训练神经网络，15％的样本数据用作验证神经网络，15％的样本数据用作测试神经网络。Divide the sample data into three parts, where 70% of the sample data is used for training the neural network, 15% of the sample data is used for the verification of the neural network, and 15% of the sample data is used for testing the neural network.
步骤2，训练神经网络模型：Step 2, train the neural network model:
神经网络模型如下：The neural network model is as follows:
隐藏层的输入函数为：The input function of the hidden layer is:
隐藏层的输出函数为：hk＝f(net_hidden)；The output function of the hidden layer is: hk = f(net_hidden);
输出层的输入函数为：The input function of the output layer is:
输出层的输出函数为：Oz＝f(net_output)＝RaThe output function of the output layer is: Oz = f(net_output) = Ra
其中，J为输入层神经元的个数，为Cj,k是输入神经元和隐含神经元之间的权重，ij是归一化后的输入参数包括金刚线的线速度、Z轴切割进给速度和太阳能硅片切割液纯度，θk是隐藏节点的偏置，K为隐藏层神经元的个数，Dk,z是隐藏神经元和输出神经元之间的权重，hk是隐藏节点的输出值，是输出节点的偏置，z为输出层神经元个数，f是tansig传递函数Among them, J is the number of neurons in the input layer, is Cj, k is the weight between the input neuron and the hidden neuron, and ij is the input parameter after normalization, including the linear velocity of the diamond wire, the Z axis Cutting feed speed and purity of solar wafer cutting fluid, θk is the bias of hidden nodes, K is the number of neurons in the hidden layer, Dk, z is the weight between hidden neurons and output neurons, hk is the output value of the hidden node, is the bias of the output node, z is the number of neurons in the output layer, f is the tansig transfer function
由于切割参数的物理量不同，各个数据之间的衡量方法不同，为解决输入参数间的数值问题、避免神经元输出饱和、加快网络的收敛速度和提高网络性能，因此样本数据在输入神经网络之前，我们需要将样本数据归一化，所用归一化的公式为：Due to the different physical quantities of the cutting parameters, the measurement methods of each data are different. In order to solve the numerical problems between the input parameters, avoid the output saturation of neurons, speed up the convergence speed of the network and improve the network performance, the sample data is input into the neural network. We need to normalize the sample data, the normalization formula used is:
其中，ij是归一化后的切割参数数据，dmax是切割参数样本数据的最大值， dmin是切割参数样本数据的最小值，dj是第j个切割参数样本数据。Among them, ij is the normalized cutting parameter data, dmax is the maximum value of the cutting parameter sample data, dmin is the minimum value of the cutting parameter sample data, and dj is the jth cutting parameter sample data.
归一化后的样本数据用于输入到如图2所示的具有三层结构的前馈式神经网络，进行神经网络训练。该神经网络的输入层节点个数为3个，隐藏层节点个数为10个，输出层节点个数为1个，w为该神经元的权重参数，b为偏置参数，激励函数为双曲正切型。其中，输入数据为线速度、Z轴速度和切割液纯度，输出数据为硅片表面粗糙度Ra。The normalized sample data is input to the feed-forward neural network with a three-layer structure as shown in Figure 2 for neural network training. The number of input layer nodes of the neural network is 3, the number of hidden layer nodes is 10, the number of output layer nodes is 1, w is the weight parameter of the neuron, b is the bias parameter, and the activation function is double Tangent type. Among them, the input data is the linear speed, the Z-axis speed and the purity of the cutting fluid, and the output data is the surface roughness Ra of the silicon wafer.
神经网络的训练流程如图3所示，输入归一化的样本数据进行神经网络训练，仿真数据并与测试数据对比，计算期望输出与实际输出之间的误差。若误差小于规定范围，则认定神经网络模型训练完成，否则进行反向传播计算隐层误差并更新权值，直至误差满足规定条件。重复循环多次训练以达到神经元之间连接权值的最优配比。The training process of the neural network is shown in Figure 3. The normalized sample data is input for neural network training, the simulated data is compared with the test data, and the error between the expected output and the actual output is calculated. If the error is less than the specified range, it is considered that the training of the neural network model is completed, otherwise, the backpropagation is performed to calculate the hidden layer error and update the weight until the error meets the specified conditions. Repeat the training cycle multiple times to achieve the optimal ratio of connection weights between neurons.
使用MATLAB软件的ANN工具箱软件包对神经网络进行仿真，训练参数选择如下：The ANN toolbox software package of MATLAB software is used to simulate the neural network, and the training parameters are selected as follows:
训练函数选用Trainlm，传递函数选用Tansig，学习函数选用Learngdm，性能评价函数选用均方误差，最大训练次数选择为1000，ANN的学习率为0.01，最初的权重和偏置由MATLAB自动导出。The training function is Trainlm, the transfer function is Tansig, the learning function is Learngdm, the performance evaluation function is mean square error, the maximum number of training times is 1000, the learning rate of ANN is 0.01, and the initial weight and bias are automatically derived by MATLAB.
步骤3，预测切割参数，对比误差并找到最优参数配置。Step 3, predict the cutting parameters, compare the errors and find the optimal parameter configuration.
使用步骤2训练完成的神经网络预测模型对所有的样本数据进行预测，表 2则显示了28组样本数据采集时所用切割参数条件下的所有预测值并计算出了平均测量值的预测结果和实际测量值之间的相对误差。表格中分别列出了线速度、Z轴速度、切割液的纯度、每个切割过程的持续时间、Ra测量值和Ra预测值。Use the neural network prediction model trained in step 2 to predict all sample data. Table 2 shows all the predicted values under the cutting parameters used in the collection of 28 groups of sample data and calculated the predicted results and actual values of the average measured values. The relative error between measurements. The table lists line speed, Z-axis speed, purity of the cutting fluid, duration of each cutting process, measured Ra and predicted Ra.
相对误差的计算公式为：The formula for calculating the relative error is:
其中实验结果为步骤1中所测实验数据，预测结果为神经网络的预测数据。The experimental result is the experimental data measured in step 1, and the predicted result is the predicted data of the neural network.
从表中数据可以看到Ra的测量值和预测值具有一定的相似性。就单行而言，部分预测数据几乎等同于测量值，超过一半的相对误差小于1％。28组数据中最小的相对误差为0.009％，最大的相对误差为5.942％，因此可以判定该预测模型具有较好的预测能力，可以用于金刚石线锯切割机的切割参数预测。 From the data in the table, it can be seen that the measured and predicted values of Ra have a certain similarity. For individual rows, some of the predicted data are nearly identical to the measured values, with more than half having a relative error of less than 1%. The smallest relative error among the 28 sets of data is 0.009%, and the largest relative error is 5.942%. Therefore, it can be judged that the prediction model has good predictive ability and can be used for cutting parameter prediction of diamond wire saw cutting machine.
依据实际经验，切割液与蒸馏水的体积比在3：7时切割效果具有显著的优势。当切割液的纯度参数选择为0.3，使用训练完成的预测模型进行切割参数预测，根据预测结果可以绘制出图4所示的折线图。其中图中的每条曲线上有着自己对应的符号标识，不同的符号标志代表着和图表右侧对应的Z轴速度。从中可以观察到一个显著的特点，在线速度相对较慢时Z轴速度对Ra值有着极大的影响。随着线速度的增加Ra的值越来越小，而原本具有劣势的1.0mm/min 的Z轴速度反而可以加工出较低Ra值的硅片。使用1mm/min的Z轴速度进行硅片加工需要耗费100分钟，浪费大量的电力和切割液并且影响硅片的生产效率，因此本组预测数据表面粗糙度Ra的需求为1.4时对应的切割参数最优解为：线速度3.6m/s，Z轴速度2mm/min。当Ra需求为1.3时，最优切割参数为：线速度5.2m/s，Z轴速度2mm/min。在实际切割操作时，切割人员可根据实际所需硅片的粗糙度选择最合适参数，以达到高效和经济的目的。According to practical experience, the cutting effect has a significant advantage when the volume ratio of cutting fluid to distilled water is 3:7. When the purity parameter of the cutting fluid is selected as 0.3, use the trained prediction model to predict the cutting parameters. According to the prediction results, the line graph shown in Figure 4 can be drawn. Each curve in the figure has its own corresponding symbol mark, and different symbol marks represent the Z-axis speed corresponding to the right side of the chart. It can be observed that a remarkable feature is that the Z-axis speed has a great influence on the Ra value when the line speed is relatively slow. With the increase of the line speed, the value of Ra becomes smaller and smaller, while the original disadvantageous Z-axis speed of 1.0mm/min can process silicon wafers with lower Ra value. It takes 100 minutes to process silicon wafers at a Z-axis speed of 1 mm/min, which wastes a lot of electricity and cutting fluid and affects the production efficiency of silicon wafers. Therefore, the cutting parameters corresponding to the data surface roughness Ra required by this group is 1.4 The optimal solution is: line speed 3.6m/s, Z-axis speed 2mm/min. When the Ra requirement is 1.3, the optimal cutting parameters are: line speed 5.2m/s, Z-axis speed 2mm/min. During the actual cutting operation, the cutting personnel can choose the most suitable parameters according to the roughness of the actual required silicon wafer, so as to achieve the purpose of high efficiency and economy.
综上所述，本发明基于神经网络的硅片切割参数预测方法能够根据少量样本数据建立神经网络预测模型，建立的模型能够根据新的切割参数预测出切割机加工后硅片的表面粗糙度。根据预测结果使得研究人员能够在考虑低Ra和低能耗的情况下选择最优的切割参数。In summary, the neural network-based silicon wafer cutting parameter prediction method of the present invention can establish a neural network prediction model based on a small amount of sample data, and the established model can predict the surface roughness of the silicon wafer after cutting according to the new cutting parameters. Based on the predicted results, the researchers were able to choose the optimal cutting parameters considering low Ra and low energy consumption.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810651839.XACN108985452A (en) | 2018-06-22 | 2018-06-22 | Silicon chip of solar cell cutting parameter prediction technique neural network based |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810651839.XACN108985452A (en) | 2018-06-22 | 2018-06-22 | Silicon chip of solar cell cutting parameter prediction technique neural network based |
| Publication Number | Publication Date |
|---|---|
| CN108985452Atrue CN108985452A (en) | 2018-12-11 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810651839.XAPendingCN108985452A (en) | 2018-06-22 | 2018-06-22 | Silicon chip of solar cell cutting parameter prediction technique neural network based |
| Country | Link |
|---|---|
| CN (1) | CN108985452A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111625997A (en)* | 2020-05-27 | 2020-09-04 | 山东天岳先进材料科技有限公司 | Crystal cutting process optimization method and device |
| CN115809523A (en)* | 2022-12-07 | 2023-03-17 | 阳江市中乌巴顿技术研究院 | Cutting parameter prediction processing method and device, electronic equipment and storage medium |
| CN117634101A (en)* | 2024-01-26 | 2024-03-01 | 杭州广立微电子股份有限公司 | Chip surface morphology determination method, chip surface morphology determination device, computer device and storage medium |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050251284A1 (en)* | 2002-03-27 | 2005-11-10 | Joze Balic | CNC control unit with learning ability for machining centers |
| CN106447092A (en)* | 2016-09-12 | 2017-02-22 | 浙江工业大学 | Marine reverse osmosis desalination system performance prediction method based on MEA-BP neural network |
| CN107199397A (en)* | 2016-03-17 | 2017-09-26 | 发那科株式会社 | Machine learning device, laser-processing system and machine learning method |
| CN107976956A (en)* | 2017-11-22 | 2018-05-01 | 沈阳机床股份有限公司 | The multi objective cutting data generating algorithm and cutting parameter optimization method of numerically-controlled machine tool |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050251284A1 (en)* | 2002-03-27 | 2005-11-10 | Joze Balic | CNC control unit with learning ability for machining centers |
| CN107199397A (en)* | 2016-03-17 | 2017-09-26 | 发那科株式会社 | Machine learning device, laser-processing system and machine learning method |
| CN106447092A (en)* | 2016-09-12 | 2017-02-22 | 浙江工业大学 | Marine reverse osmosis desalination system performance prediction method based on MEA-BP neural network |
| CN107976956A (en)* | 2017-11-22 | 2018-05-01 | 沈阳机床股份有限公司 | The multi objective cutting data generating algorithm and cutting parameter optimization method of numerically-controlled machine tool |
| Title |
|---|
| 刘月云: "基于BP神经网络的电火花线切割多次切割工艺参数的预测研究", 《模具制造》* |
| 吕智等: "《超硬材料工具设计与制造》", 31 January 2010, 北京冶金工业出版社* |
| 庞桂兵等: "基于神经网络的电化学加工表面粗糙度预测与加工参数正交优化", 《中国机械工程》* |
| 董成祥等: "硅片自旋转磨削表面粗糙度建模与分析", 《金刚石与磨料磨具工程》* |
| 顾丰: "电火花微小孔加工工艺参数优化及建模的研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技I辑》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111625997A (en)* | 2020-05-27 | 2020-09-04 | 山东天岳先进材料科技有限公司 | Crystal cutting process optimization method and device |
| CN111625997B (en)* | 2020-05-27 | 2024-01-19 | 山东天岳先进科技股份有限公司 | Crystal cutting process optimization method and device |
| CN115809523A (en)* | 2022-12-07 | 2023-03-17 | 阳江市中乌巴顿技术研究院 | Cutting parameter prediction processing method and device, electronic equipment and storage medium |
| CN117634101A (en)* | 2024-01-26 | 2024-03-01 | 杭州广立微电子股份有限公司 | Chip surface morphology determination method, chip surface morphology determination device, computer device and storage medium |
| Publication | Publication Date | Title |
|---|---|---|
| CN111369070B (en) | Multimode fusion photovoltaic power prediction method based on envelope clustering | |
| CN114564880B (en) | A digital twin module construction method for additive manufacturing process | |
| WO2021174525A1 (en) | Parts surface roughness and cutting tool wear prediction method based on multi-task learning | |
| WO2021128576A1 (en) | Tool condition monitoring dataset enhancement method based on generative adversarial network | |
| CN108985452A (en) | Silicon chip of solar cell cutting parameter prediction technique neural network based | |
| CN106779069A (en) | A kind of abnormal electricity consumption detection method based on neutral net | |
| CN107292051A (en) | A kind of carbide chip chemically-mechanicapolish polishes the Forecasting Methodology of surface roughness | |
| CN103761429A (en) | Milling workpiece surface roughness predicting method | |
| CN114109859A (en) | A Neural Network Prediction Method of Centrifugal Pump Performance Without Flow Sensing | |
| CN108764473A (en) | A kind of BP neural network water demands forecasting method based on correlation analysis | |
| CN106777775A (en) | A kind of neural net method based on many section water level forecast river discharges | |
| CN113189948B (en) | Optimization method of machining process parameters for thin plate parts considering reliability of machining accuracy | |
| CN103279803A (en) | Load modeling method and system based on comprehensive information theory and modern interior point theory | |
| CN112182709A (en) | Rapid prediction method for let-down water temperature of large-scale reservoir stop log door layered water taking facility | |
| CN114290503B (en) | Power control method and device for T-beam steam health generator | |
| CN114966409B (en) | A method for estimating the state of charge of a power lithium battery based on a multi-layer perceptron algorithm | |
| CN112528955B (en) | High-frequency element machining size precision prediction method and system | |
| CN103793577A (en) | Method for optimizing and controlling little-cutting fluid processing technology in machining process | |
| CN112330045A (en) | A Loss Evaluation and Loss Reduction Method of Transmission Network Based on K-medoids Cluster Analysis | |
| CN111948977A (en) | Multi-objective optimization method and system for stainless steel processing | |
| CN114936742A (en) | Water supply system scheduling agent decision method | |
| CN116933631A (en) | Blasting block prediction method based on neural network optimization algorithm | |
| CN116415191A (en) | A SSAE-Based Online Prediction Method of Surface Roughness under Multiple Working Conditions in NC Milling | |
| CN110059443B (en) | A Rapid Prediction Method for Intake and Release Water Temperature of Layered Reservoir | |
| CN101118569A (en) | Method for Determining Injection Time Parameters in Plastic Injection Molding Process |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20181211 |