技术领域technical field
本发明属于数控机床刀具磨损监测技术领域，更具体的说，涉及一种基于多传感器信息融合及深度置信网络的刀具磨损监测方法。The invention belongs to the technical field of tool wear monitoring of numerically controlled machine tools, and more particularly relates to a tool wear monitoring method based on multi-sensor information fusion and a deep confidence network.
技术背景technical background
刀具磨损是影响加工工业中工件质量的关键因素，有效且准确地预测刀具磨损可以使刀具及时更换，以避免不必要的浪费；研究表明，CNC机床配备刀具监测系统后可减少故障停机时间的75％，生产效率提高10-60％，机床利用率提高50％，稳定、准确的刀具监测系统是现代化加工必不可少的。Tool wear is a key factor affecting the quality of workpieces in the machining industry. Effective and accurate prediction of tool wear can enable timely tool replacement to avoid unnecessary waste; studies have shown that CNC machine tools can be equipped with tool monitoring systems to reduce downtime by 75%. %, the production efficiency is increased by 10-60%, the machine tool utilization rate is increased by 50%, and a stable and accurate tool monitoring system is essential for modern processing.
在监测刀具状态时，根据工作场合会采用多种传感器来监测刀具的使用状况，现有的方法通常是利用单一传感器信号的某种特征参数来表示刀具的磨损状态，监测的准确性受限于某一传感器的精度，监测稳定性较差，不能有效地实现刀具状态的监测。When monitoring the tool status, a variety of sensors are used to monitor the tool usage according to the workplace. The existing method usually uses a certain characteristic parameter of a single sensor signal to indicate the tool wear status. The accuracy of monitoring is limited by The accuracy of a certain sensor is poor, and the monitoring stability is poor, so it cannot effectively monitor the tool state.
当下制造业中，对传感器信息进行预处理、特征提取、特征选择主要依赖于技术人员的信号处理技术和诊断经验，远远达不到智能化的要求，本领域需要一种稳定、精准的智能化刀具磨损监测系统。In the current manufacturing industry, the preprocessing, feature extraction, and feature selection of sensor information mainly rely on the signal processing technology and diagnosis experience of technicians, which are far from meeting the requirements of intelligence. The field needs a stable and accurate intelligent Chemical tool wear monitoring system.
发明内容SUMMARY OF THE INVENTION
要解决的技术问题：Technical problem to be solved:
针对现有技术的缺陷或改进需求,本发明提出了一种基于信息融合及深度置信网络的刀具磨损监测方法，其基于现有的刀具状态监测特点，研究及设计了一种一种基于信息融合及深度置信网络的刀具磨损监测方法；所述监测方法融合了多种传感器信号的有效信息，且利用深度置信网络在线识别出刀具的磨损状态，实现了对刀具状态的稳定在线监测；该方法能够摆脱对信号处理技术和诊断经验的依赖，实现刀具磨损特征的自适应提取，并且提高了监测准确性及灵活性，监测不再受限于某一个传感器信号。Aiming at the defects or improvement needs of the prior art, the present invention proposes a tool wear monitoring method based on information fusion and deep confidence network. and a tool wear monitoring method based on a deep belief network; the monitoring method integrates the effective information of various sensor signals, and uses the deep belief network to identify the wear state of the tool online, and realizes the stable online monitoring of the tool state; this method can Get rid of the dependence on signal processing technology and diagnosis experience, realize the self-adaptive extraction of tool wear characteristics, and improve the monitoring accuracy and flexibility, and monitoring is no longer limited by a certain sensor signal.
技术方案：Technical solutions:
步骤1：在某一工况下，使用恒定的切削参数对材料进行加工，刀具在侧边上铣削加工，测量该过程中的切削力信号、振动信号，同时测量每次加工后刀具的后刀面磨损量，并将归一化处理后的刀面磨损值作为神经网络的输出值；Step 1: Under a certain working condition, use constant cutting parameters to process the material, the tool is milled on the side, measure the cutting force signal and vibration signal in the process, and measure the flank of the tool after each machining. face wear amount, and take the normalized face wear value as the output value of the neural network;
步骤2：分别提取每一种传感器信号在时域、频域及时频域上的特征参数，并进行归一化处理；Step 2: Extract the characteristic parameters of each sensor signal in the time domain, frequency domain and frequency domain respectively, and perform normalization processing;
步骤3：将归一化处理后的特征参数作为深度置信网络(DBN)的输入向量对识别模型进行训练，在最后一层加softmax分类器，通过BP算法进行微调，最终输出刀具磨损状态。Step 3: Use the normalized feature parameters as the input vector of the Deep Belief Network (DBN) to train the recognition model, add a softmax classifier to the last layer, fine-tune it through the BP algorithm, and finally output the tool wear state.
进一步地，所述传感器信号包括振动信号、及切削力信号。Further, the sensor signal includes a vibration signal and a cutting force signal.
进一步地，提取的传感器信号在时域的特征参数包括均值、均方值、方根均值、绝对均值、绝对值总和、极大值、自成均值、波高率、波形率、标准差、歪度。Further, the characteristic parameters of the extracted sensor signal in the time domain include mean value, mean square value, root mean value, absolute mean value, absolute value sum, maximum value, self-generated mean value, wave height rate, wave rate, standard deviation, skewness .
进一步地，提取的传感器信号在时域的特征参数包括功率谱均值、功率谱偏斜度、功率谱峭度、功率谱脉峰值、功率谱方差、功率谱脉冲指标。Further, the extracted characteristic parameters of the sensor signal in the time domain include power spectrum mean, power spectrum skewness, power spectrum kurtosis, power spectrum pulse peak value, power spectrum variance, and power spectrum pulse index.
进一步地，在频域上提取特征参数时，将振动信号通过离散傅里叶变换以得到功率谱；其中，所述重心频率、所述均方频率及所述均方根频率表示功率谱的主频带位置的变化情况，所述频率方差及所述频率标准差表示能量谱的离散程度。Further, when extracting characteristic parameters in the frequency domain, the vibration signal is subjected to discrete Fourier transform to obtain a power spectrum; wherein, the center of gravity frequency, the mean square frequency and the root mean square frequency represent the main power spectrum of the power spectrum. The variation of the frequency band position, the frequency variance and the frequency standard deviation represent the degree of dispersion of the energy spectrum.
进一步地，在时频域特征提取时，使用db5三层小波包分解对切削力信号、振动信号进行分解重构，提取其8个小波包能量带作为时频特征。Further, in the time-frequency domain feature extraction, the db5 three-layer wavelet packet decomposition is used to decompose and reconstruct the cutting force signal and the vibration signal, and its 8 wavelet packet energy bands are extracted as time-frequency features.
进一步地，采用Min-Max Normalization对提取到的特征参数做归一化处理，公式为：Further, Min-Max Normalization is used to normalize the extracted feature parameters, and the formula is:
式中，x为输入值；y为归一化输出值；Min为最小值；Max为最大值。In the formula, x is the input value; y is the normalized output value; Min is the minimum value; Max is the maximum value.
进一步地，深度置信网络(DBN)由若干层神经元构成，组成元件是受限玻尔兹曼机(RBM)，其训练过程包括预训练、微调和预测。Further, the deep belief network (DBN) is composed of several layers of neurons, and the constituent element is a restricted Boltzmann machine (RBM), and its training process includes pre-training, fine-tuning and prediction.
进一步地，受限玻尔兹曼机(RBM)中,设定显层有nv个神经元，隐藏层有nh个神经元，v＝(v1,v2,v3…,vnv)T是可见层状态向量(振动传感器和力传感器信息的特征参数)，隐层状态向量为h＝(h1,h2,h3,...,hnh)T，有vi，hj∈{0，1}，为显曾的偏置向量，为隐层的偏置向量，W＝(wi,j)为隐层和显层之间的权值矩阵，记θ＝(W,a,b)，每一个神经元被激活的概率如下式所示：Further, in the restricted Boltzmann machine (RBM), it is assumed that there are nv neurons in the display layer and nh neurons in the hidden layer, v=(v1 , v2 , v3 ..., vnv )T is the visible layer state vector (feature parameters of vibration sensor and force sensor information), and the hidden layer state vector is h=(h1 , h2 , h3 ,...,hnh )T , With vi , hj ∈ {0, 1}, is the bias vector of Xian Zeng, is the bias vector of the hidden layer, W=(wi,j ) is the weight matrix between the hidden layer and the visible layer, denoted θ=(W,a,b), the probability of each neuron being activated is as follows shown:
对于训练样本其中ns为S组切削过程中传感器信息的特征参数，其中S为特征提取出刀具切削过程中传感器的特征参数，训练RBM的目的就是最大化如下的似然函数：for training samples where ns is the characteristic parameter of the sensor information during the cutting process of the S group, Among them, S is the feature to extract the feature parameters of the sensor during the cutting process of the tool. The purpose of training the RBM is to maximize the following likelihood function:
最大化公式(1)的目的是为了找到最佳参数θ，可以对其负数进行随机梯度下降法来确定该值：The purpose of maximizing formula (1) is to find the optimal parameter θ, which can be determined by performing stochastic gradient descent on its negative numbers:
<·>data表示对数据分布的数学期望，<·>model表示对模型分布的数学期望，<·>model的计算复杂程度是运用Gibbs采样方法进行采样时，用样本<·>model对进行估计，需要大量的样本才能满足精度，大大的加大了RBM的复杂程度；本发明运用对比散度算法(Contrastive Divergence,CD)结合Gibbs采样来训练RBM，Gibbs采样从给定的可见层数据(v0)，开始计算隐藏层神经元的初始值(h0)，再通过(h0)来计算(v1)，如此循环执行K次Gibbs采样，理论上K趋近无穷时可以获得<·>model的准确值，然而在实践中仅仅几步就可以满足需求。<·>data represents the mathematical expectation of the data distribution, <·>model represents the mathematical expectation of the model distribution, and the computational complexity of the <·>model is When the Gibbs sampling method is used for sampling, the sample <·>model pair is used to estimate, which requires a large number of samples to meet the accuracy, which greatly increases the complexity of the RBM; the present invention uses the Contrastive Divergence algorithm (Contrastive Divergence, CD) combined with Gibbs sampling to train RBM, Gibbs sampling starts from the given visible layer data (v0 ), starts to calculate the initial value of the hidden layer neurons (h0 ), and then calculates (v1 ) through (h0 ), and so on. K times of Gibbs sampling, theoretically, when K approaches infinity, the exact value of the <·>model can be obtained, but in practice, only a few steps can meet the requirements.
有益效果：Beneficial effects:
本发明提出了一种基于信息融合及深度置信网络的刀具磨损监测方法，采集数控机床的多种传感器信号，在时域、频域及时频域上的特征参数，并对提取到的特征参数做归一化处理，提高了灵活性及监测准确性；DBN网络克服了传统刀具监测技术对信号处理技术和诊断经验的依赖，能够自适应的提取可以表征刀具磨损状态的传感器信息特征无须依靠专家经验，运算速度快，最大程度上减少人工特征对结果的影响，具有多层次的特征表达能力，获取更加抽象的数据特征，提高数据特征的有效性。The invention proposes a tool wear monitoring method based on information fusion and deep confidence network, collects various sensor signals of numerically controlled machine tools, characteristic parameters in time domain, frequency domain and frequency domain, and performs the extraction of characteristic parameters. Normalized processing improves flexibility and monitoring accuracy; DBN network overcomes the dependence of traditional tool monitoring technology on signal processing technology and diagnostic experience, and can adaptively extract sensor information features that can characterize tool wear status without relying on expert experience , the operation speed is fast, the influence of artificial features on the results is minimized, and it has the ability to express multi-level features, obtain more abstract data features, and improve the effectiveness of data features.
附图说明Description of drawings
图1为刀具磨损在线监测方法流程图。Fig. 1 is the flow chart of the online monitoring method of tool wear.
图2为小波包分解结构图。Fig. 2 is a structure diagram of wavelet packet decomposition.
图3为RBM网络结构。Figure 3 shows the RBM network structure.
图4DBN神经网络预训练过程。Figure 4. DBN neural network pre-training process.
图5为cd-k算法。Figure 5 shows the cd-k algorithm.
图6为DBN、ANN、SVM训练误差对比图。Figure 6 is a comparison chart of DBN, ANN, and SVM training errors.
图7为DBN、ANN、SVM训练时间对比图。Figure 7 is a comparison diagram of the training time of DBN, ANN, and SVM.
具体实施方式Detailed ways
现结合实施例、附图对本发明作进一步描述:Now in conjunction with embodiment, accompanying drawing the present invention is further described:
本发明提出一种基于多传感器信息融合及深度置信网络的刀具磨损监测方法，附图1表示该刀具磨损监测方法具体流程，主要包括三个步骤：The present invention proposes a tool wear monitoring method based on multi-sensor information fusion and deep confidence network. Figure 1 shows the specific process of the tool wear monitoring method, which mainly includes three steps:
步骤1:在某一工况下，使用恒定的切削参数对零件进行加工，铣刀在工件上铣削加工径向距离1mm，轴向切深2mm的侧边70次，刀具从初始磨损变为磨钝磨损，测量每次加工后刀具的后刀面磨损量，Kistler加速度传感器采取三个方向的振动信号，同时采用KIStler9123C旋转式测力计测量该过程中三个方向的切削力，提取其中50组刀具磨损测量值作为DNN神经网络的输出值进行训练；使用恒定切削参数可以减少变形量，降低计算难度，并且旋转测力仪具有测力信号稳定，安装方便，抗干扰能力强等优点。Step 1: Under a certain working condition, use constant cutting parameters to process the part. The milling cutter mills the side of the workpiece with a radial distance of 1mm and an axial depth of cut 2mm 70 times. The tool changes from initial wear to grinding. Dull wear, measure the flank wear of the tool after each machining, the Kistler accelerometer takes vibration signals in three directions, and uses the KIStler9123C rotary dynamometer to measure the cutting force in three directions during the process, and extracts 50 sets of them. The measured value of tool wear is used as the output value of the DNN neural network for training; the use of constant cutting parameters can reduce the amount of deformation and reduce the difficulty of calculation, and the rotary dynamometer has the advantages of stable force measurement signal, convenient installation, and strong anti-interference ability.
步骤2：分别提取每一种传感器信号在时域、频域及时频域上的特征参数，并进行归一化处理。Step 2: Extract the characteristic parameters of each sensor signal in the time domain, frequency domain, and frequency domain respectively, and perform normalization processing.
具体地，采集多种传感器信号，对每一种传感器信号分别在时域、频域及时频域上提取特征参数，并将提取后的所有特征参数做归一化处理。本实施方式中，在时域提取的特征参数包括均值、均方值、方根均值、绝对均值、绝对值总和、极大值、自成均值、波高率、波形率、标准差、歪度；在时域提取特征参数时采用基于自适应噪声的完备经验模态分解方法(CEEMDAN)，提取的时域特征为模态函数的能量值。Specifically, a variety of sensor signals are collected, characteristic parameters are extracted for each sensor signal in the time domain, frequency domain, and frequency domain, respectively, and all the extracted characteristic parameters are normalized. In this embodiment, the characteristic parameters extracted in the time domain include mean value, mean square value, root mean value, absolute mean value, absolute value sum, maximum value, self-generated mean value, wave height rate, wave rate, standard deviation, and skewness; The complete empirical mode decomposition method based on adaptive noise (CEEMDAN) is used to extract the feature parameters in the time domain, and the extracted time domain features are the energy values of the modal function.
其中,(1)均值:Among them, (1) mean:
(2)均方值：(2) Mean square value:
(3)方根均值：(3) mean square root:
(4)绝对值总和：(4) Sum of absolute values:
(5)极大值：(5) Maximum value:
max(Si)max(Si )
(6)自成均值：(6) Self-contained mean:
(7)波高率：(7) Wave height rate:
(8)波形率：(8) Wave rate:
(9)标准差：(9) Standard deviation:
(10)歪度：(10) skewness:
(11)绝对均值：(11) Absolute mean:
在频域上提取特征时，提取每个传感器以下六个特征参数作为频域特征。When extracting features in the frequency domain, the following six feature parameters of each sensor are extracted as frequency domain features.
(1)功率谱均值：(1) Average power spectrum:
(2)功率谱偏斜度：(2) Power spectrum skewness:
(3)功率谱峭度：(3) Power spectrum kurtosis:
(4)功率谱脉峰值：(4) Peak value of power spectrum pulse:
max(S(f))max(S(f))
(5)功率谱方差：(5) Power spectrum variance:
(6)功率谱脉冲指标：(6) Power spectrum pulse index:
在时频域特征提取时，使用db5三层小波包分解对切削力信号、振动信号进行分解重构:In the time-frequency domain feature extraction, the db5 three-layer wavelet packet decomposition is used to decompose and reconstruct the cutting force signal and vibration signal:
其中，n为频率指标，k为位置指标，j为尺度指标，wn称为关于ψ(t)的正交小波包基，由一个标准正交化的尺度函数ψ(t)，w0＝ψ(t)，由双尺度差分递归方程组，生成函数组：Among them, n is the frequency index, k is the position index, j is the scale index,wn is called the orthogonal wavelet packet basis about ψ(t), and it is a standard orthogonalized scale function ψ(t), w0 = ψ(t), from the system of two-scale difference recursive equations, generates a set of functions:
其中,hk,gk为ψ(t)导出的一对共轭正交滤波器系数。Among them, hk , gk are a pair of conjugate orthogonal filter coefficients derived from ψ(t).
小波包可以随着分辨率2j的增加，其变宽的频谱窗口具有进一步分割变细的优良品质，对于给定的信号，通过一组低高通组合，正交滤波器hk、gk可以将信号划分至任意频段上，使其在低频和高频都具有较高的时间和频率分辨率；本发明专利首先通过对加工过程中的切削力信号进行三层小波包分解，从小波包系数重构图中提取与磨损相关的切削力特征和切削振动特征，小波包分解过程如附图2的结构进行划分，j表示分解层数；其作用是:三层小波包分解可以将采样频率8kHZ的切削力信号，在频域上细分为8段，在时域上将每个频段的小波包系数进行重构，并将此信号的能量值作为刀具磨损特征值。With the increase of the resolution 2j, the wavelet packet has the good quality of further segmentation and thinning in the wider spectral window. For a given signal, through a set of low-high-pass combinations, the quadrature filters hk and gk can divide the signal into to any frequency band, so that it has high time and frequency resolution at both low and high frequencies; the patent of the present invention firstly decomposes the cutting force signal in the machining process by three layers of wavelet packets, and reconstructs the graph from the wavelet packet coefficients. Extract the cutting force feature and the cutting vibration feature relevant to wear in, the wavelet packet decomposition process is divided as the structure of accompanying drawing 2, and j represents the decomposition layer number; The signal is subdivided into 8 segments in the frequency domain, the wavelet packet coefficients of each frequency band are reconstructed in the time domain, and the energy value of this signal is used as the tool wear characteristic value.
本实施方式中，采用Min-Max Normalization对提取到的特征参数做归一化处理，公式为:In the present embodiment, Min-Max Normalization is adopted to normalize the extracted feature parameters, and the formula is:
式中，x为输入值；y为归一化输出值；Min为最小值；Max为最大值。In the formula, x is the input value; y is the normalized output value; Min is the minimum value; Max is the maximum value.
步骤3：将从步骤2得到的50组特征作为DNN网络的输入，将测量的50组刀具后刀面磨损状态作为DNN神经网络的输出端对网络进行训练，将其余的20组作为刀具磨损监测实验；输入的切削力、振动特征共150个，经过DBN神经网络进行特征提取，图3为RBM网络构图；利用DBN神经网络，逐层提取能够表征刀具模塑的特征，完成预训练，图4为预训练过程图；然后通过BP算法微调整个网络参数，最终输出刀具磨损状态。Step 3: Use the 50 sets of features obtained from Step 2 as the input of the DNN network, use the measured 50 sets of tool flank wear states as the output of the DNN neural network to train the network, and use the remaining 20 sets as tool wear monitoring Experiment: There are 150 input cutting force and vibration features, which are extracted by DBN neural network. Figure 3 shows the composition of RBM network; using DBN neural network, features that can characterize tool molding are extracted layer by layer, and pre-training is completed, Figure 4 It is the pre-training process diagram; then fine-tune the entire network parameters through the BP algorithm, and finally output the tool wear state.
受限玻尔兹曼机(RBM)中,设定显层有nv个神经元，隐藏层有nh个神经元，是可见层状态向量(振动传感器和力传感器信息的特征参数)，隐层状态向量为h＝(h1,h2,h3,...,hnh)T，有vi,hj∈{0,1},为显曾的偏置向量，为隐层的偏置向量，W＝(wi,j)为隐层和显层之间的权值矩阵，记θ＝(W,a,b)，每一个神经元被激活的概率如下式所示：In Restricted Boltzmann Machine (RBM), it is assumed that there are nv neurons in the display layer and nh neurons in the hidden layer. is the visible layer state vector (feature parameter of vibration sensor and force sensor information), the hidden layer state vector is h=(h1 , h2 , h3 ,...,hnh )T , With vi ,hj ∈{0,1}, is the bias vector of Xian Zeng, is the bias vector of the hidden layer, W=(wi,j ) is the weight matrix between the hidden layer and the visible layer, denoted θ=(W,a,b), the probability of each neuron being activated is as follows shown:
对于训练样本其中ns为S组切削过程中传感器信息的特征参数，其中S为特征提取出刀具切削过程中传感器的特征参数，训练RBM的目的就是最大化如下的似然函数：for training samples where ns is the characteristic parameter of the sensor information during the cutting process of the S group, Among them, S is the feature to extract the feature parameters of the sensor during the cutting process of the tool. The purpose of training the RBM is to maximize the following likelihood function:
最大化公式(1)的目的是为了找到最佳参数θ，可以对其负数进行随机梯度下降法来确定该值：The purpose of maximizing formula (1) is to find the optimal parameter θ, which can be determined by performing stochastic gradient descent on its negative numbers:
<·>data表示对数据分布的数学期望，<·>model表示对模型分布的数学期望，<·>model的计算复杂程度是运用Gibbs采样方法进行采样时，用样本<·>model对进行估计，需要大量的样本才能满足精度，大大的加大了RBM的复杂程度；本发明运用对比散度算法(Contrastive Divergence,CD)结合Gibbs采样来训练RBM，如附图5所示，Gibbs采样从给定的可见层数据(v0)，开始计算隐藏层神经元的初始值(h0)，再通过(h0)来计算(v1)，如此循环执行3次Gibbs采样，理论上K趋近无穷时可以获得<·>model的准确值，然而在实践中仅仅3步就可以满足需求。<·>data represents the mathematical expectation of the data distribution, <·>model represents the mathematical expectation of the model distribution, and the computational complexity of the <·>model is When the Gibbs sampling method is used for sampling, the sample <·>model pair is used to estimate, which requires a large number of samples to meet the accuracy, which greatly increases the complexity of the RBM; the present invention uses the Contrastive Divergence algorithm (Contrastive Divergence, CD) combined with Gibbs sampling to train RBM, as shown in Figure 5, Gibbs sampling starts from the given visible layer data (v0 ), starts to calculate the initial value of hidden layer neurons (h0 ), and then calculates (h0 ) through (h 0 ) ( v1 ), 3 times of Gibbs sampling is performed in this way. In theory, when K approaches infinity, the exact value of the <·>model can be obtained. However, in practice, only 3 steps can meet the requirements.
结合附图6，附图7，采用两种不同的监测方法支持向量机(SVM)和人工神经网络(ANN网络)与本发明一种基于多传感器信息融合及深度置信网络的刀具磨损监测方法进行对比验证，对比结果显示本发明提供的方法准确率和速度都较高，由此表明本方法稳定性更好，由此表明本专利稳定性更好优于其他两种方法。In conjunction with accompanying drawing 6, accompanying drawing 7, adopt two different monitoring methods Support Vector Machine (SVM) and artificial neural network (ANN network) and a kind of tool wear monitoring method based on multi-sensor information fusion and deep confidence network of the present invention to carry out. By comparison and verification, the comparison results show that the method provided by the present invention has higher accuracy and speed, thus indicating that the method has better stability, thus indicating that the stability of the patent is better than the other two methods.
综上所述，本发明提出的刀具磨损监测方法可以自适应的提取传感器中刀具磨损状况信息，摆脱了对大量信号处理知识和诊断工程经验的依赖，并取得了较高的监测精度，和运算速度，在面对复杂的加工环境时，可以更准确的识别刀具磨损状态。To sum up, the tool wear monitoring method proposed by the present invention can adaptively extract the tool wear condition information in the sensor, get rid of the dependence on a large amount of signal processing knowledge and diagnostic engineering experience, and achieve high monitoring accuracy, and computing Speed, in the face of complex processing environment, can more accurately identify the tool wear state.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910306988.7ACN110000610A (en) | 2019-04-17 | 2019-04-17 | A kind of Tool Wear Monitoring method based on multi-sensor information fusion and depth confidence network |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910306988.7ACN110000610A (en) | 2019-04-17 | 2019-04-17 | A kind of Tool Wear Monitoring method based on multi-sensor information fusion and depth confidence network |
| Publication Number | Publication Date |
|---|---|
| CN110000610Atrue CN110000610A (en) | 2019-07-12 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910306988.7APendingCN110000610A (en) | 2019-04-17 | 2019-04-17 | A kind of Tool Wear Monitoring method based on multi-sensor information fusion and depth confidence network |
| Country | Link |
|---|---|
| CN (1) | CN110000610A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110346130A (en)* | 2019-07-19 | 2019-10-18 | 北京理工大学 | A kind of boring flutter detection method based on empirical mode decomposition and time-frequency multiple features |
| CN110561193A (en)* | 2019-09-18 | 2019-12-13 | 杭州友机技术有限公司 | Cutter wear assessment and monitoring method and system based on feature fusion |
| CN111325112A (en)* | 2020-01-31 | 2020-06-23 | 贵州大学 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
| CN111488968A (en)* | 2020-03-03 | 2020-08-04 | 国网天津市电力公司电力科学研究院 | Method and system for feature extraction of comprehensive energy metering data |
| CN111761409A (en)* | 2020-07-09 | 2020-10-13 | 内蒙古工业大学 | A deep learning-based tool wear monitoring method for multi-sensor CNC machine tools |
| CN111881860A (en)* | 2020-07-31 | 2020-11-03 | 重庆理工大学 | Modeling method of hob wear in-situ recognition model and hob wear in-situ recognition method |
| CN112192250A (en)* | 2020-10-20 | 2021-01-08 | 衢州学院 | Information acquisition and fusion device and method based on thermal design of spindle system |
| CN112365935A (en)* | 2020-10-20 | 2021-02-12 | 燕山大学 | Cement free calcium soft measurement method based on multi-scale depth network |
| WO2021174525A1 (en)* | 2020-03-06 | 2021-09-10 | 大连理工大学 | Parts surface roughness and cutting tool wear prediction method based on multi-task learning |
| CN114888635A (en)* | 2022-04-27 | 2022-08-12 | 哈尔滨理工大学 | Cutter state monitoring method |
| CN115157006A (en)* | 2022-09-08 | 2022-10-11 | 中科航迈数控软件(深圳)有限公司 | Component wear detection method and device based on information entropy, terminal and storage medium |
| CN118329196A (en)* | 2024-06-17 | 2024-07-12 | 济南红太阳数控设备有限公司 | Cutter vibration signal detection method of vibrating cutter cutting machine |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB8323284D0 (en)* | 1982-09-03 | 1983-10-05 | Valeron Corp | Worn tool detector |
| JPH10143218A (en)* | 1996-11-08 | 1998-05-29 | Nissan Motor Co Ltd | Robot cycle time prediction device |
| US20060089743A1 (en)* | 2004-10-25 | 2006-04-27 | Ford Motor Company | Data management and networking system and method |
| US7117050B2 (en)* | 2002-11-08 | 2006-10-03 | Toshiba Kikai Kabushiki Kaisha | Management supporting apparatus, management supporting system, management supporting method, management supporting program, and a recording medium with the program recorded therein |
| CN104157290A (en)* | 2014-08-19 | 2014-11-19 | 大连理工大学 | A speaker recognition method based on deep learning |
| CN105196114A (en)* | 2015-11-05 | 2015-12-30 | 西安科技大学 | Real-time online tool wear monitoring method based on wavelet analysis and neural network |
| CN106769048A (en)* | 2017-01-17 | 2017-05-31 | 苏州大学 | Adaptive deep confidence network bearing fault diagnosis method based on Nesterov momentum method |
| CN108107838A (en)* | 2017-12-27 | 2018-06-01 | 山东大学 | A kind of numerical control equipment tool wear monitoring method based on cloud knowledge base and machine learning |
| CN108710771A (en)* | 2018-05-31 | 2018-10-26 | 西安电子科技大学 | Mechanized equipment service reliability appraisal procedure based on the integrated extraction of depth characteristic |
| CN108747590A (en)* | 2018-06-28 | 2018-11-06 | 哈尔滨理工大学 | A kind of tool wear measurement method based on rumble spectrum and neural network |
| CN108830308A (en)* | 2018-05-31 | 2018-11-16 | 西安电子科技大学 | A kind of Modulation Identification method that traditional characteristic signal-based is merged with depth characteristic |
| CN109434564A (en)* | 2018-12-21 | 2019-03-08 | 哈尔滨理工大学 | A kind of cutter wear state monitoring method based on deep neural network |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB8323284D0 (en)* | 1982-09-03 | 1983-10-05 | Valeron Corp | Worn tool detector |
| JPH10143218A (en)* | 1996-11-08 | 1998-05-29 | Nissan Motor Co Ltd | Robot cycle time prediction device |
| US7117050B2 (en)* | 2002-11-08 | 2006-10-03 | Toshiba Kikai Kabushiki Kaisha | Management supporting apparatus, management supporting system, management supporting method, management supporting program, and a recording medium with the program recorded therein |
| US20060089743A1 (en)* | 2004-10-25 | 2006-04-27 | Ford Motor Company | Data management and networking system and method |
| CN104157290A (en)* | 2014-08-19 | 2014-11-19 | 大连理工大学 | A speaker recognition method based on deep learning |
| CN105196114A (en)* | 2015-11-05 | 2015-12-30 | 西安科技大学 | Real-time online tool wear monitoring method based on wavelet analysis and neural network |
| CN106769048A (en)* | 2017-01-17 | 2017-05-31 | 苏州大学 | Adaptive deep confidence network bearing fault diagnosis method based on Nesterov momentum method |
| CN108107838A (en)* | 2017-12-27 | 2018-06-01 | 山东大学 | A kind of numerical control equipment tool wear monitoring method based on cloud knowledge base and machine learning |
| CN108710771A (en)* | 2018-05-31 | 2018-10-26 | 西安电子科技大学 | Mechanized equipment service reliability appraisal procedure based on the integrated extraction of depth characteristic |
| CN108830308A (en)* | 2018-05-31 | 2018-11-16 | 西安电子科技大学 | A kind of Modulation Identification method that traditional characteristic signal-based is merged with depth characteristic |
| CN108747590A (en)* | 2018-06-28 | 2018-11-06 | 哈尔滨理工大学 | A kind of tool wear measurement method based on rumble spectrum and neural network |
| CN109434564A (en)* | 2018-12-21 | 2019-03-08 | 哈尔滨理工大学 | A kind of cutter wear state monitoring method based on deep neural network |
| Title |
|---|
| 付洋: "切削加工过程中振动状态及刀具磨损的智能监测技术研究", 《中国优秀博士学位论文全文数据库工程科技Ⅰ辑》* |
| 李威霖: "车铣刀具磨损状态监测及预测关键技术研究", 《中国优秀博士学位论文全文数据库工程科技Ⅰ辑》* |
| 林杨: "基于深度学习的刀具磨损状态监测技术的研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅰ辑》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110346130A (en)* | 2019-07-19 | 2019-10-18 | 北京理工大学 | A kind of boring flutter detection method based on empirical mode decomposition and time-frequency multiple features |
| CN110561193A (en)* | 2019-09-18 | 2019-12-13 | 杭州友机技术有限公司 | Cutter wear assessment and monitoring method and system based on feature fusion |
| CN111325112A (en)* | 2020-01-31 | 2020-06-23 | 贵州大学 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
| CN111325112B (en)* | 2020-01-31 | 2023-04-07 | 贵州大学 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
| CN111488968A (en)* | 2020-03-03 | 2020-08-04 | 国网天津市电力公司电力科学研究院 | Method and system for feature extraction of comprehensive energy metering data |
| WO2021174525A1 (en)* | 2020-03-06 | 2021-09-10 | 大连理工大学 | Parts surface roughness and cutting tool wear prediction method based on multi-task learning |
| US11761930B2 (en) | 2020-03-06 | 2023-09-19 | Dalian University Of Technology | Prediction method of part surface roughness and tool wear based on multi-task learning |
| CN111761409A (en)* | 2020-07-09 | 2020-10-13 | 内蒙古工业大学 | A deep learning-based tool wear monitoring method for multi-sensor CNC machine tools |
| CN111881860B (en)* | 2020-07-31 | 2022-05-03 | 重庆理工大学 | Modeling method of hob wear in-situ recognition model and hob wear in-situ recognition method |
| CN111881860A (en)* | 2020-07-31 | 2020-11-03 | 重庆理工大学 | Modeling method of hob wear in-situ recognition model and hob wear in-situ recognition method |
| CN112365935A (en)* | 2020-10-20 | 2021-02-12 | 燕山大学 | Cement free calcium soft measurement method based on multi-scale depth network |
| CN112365935B (en)* | 2020-10-20 | 2022-08-30 | 燕山大学 | Cement free calcium soft measurement method based on multi-scale depth network |
| CN112192250A (en)* | 2020-10-20 | 2021-01-08 | 衢州学院 | Information acquisition and fusion device and method based on thermal design of spindle system |
| CN114888635A (en)* | 2022-04-27 | 2022-08-12 | 哈尔滨理工大学 | Cutter state monitoring method |
| CN114888635B (en)* | 2022-04-27 | 2023-07-25 | 哈尔滨理工大学 | Cutter state monitoring method |
| CN115157006A (en)* | 2022-09-08 | 2022-10-11 | 中科航迈数控软件(深圳)有限公司 | Component wear detection method and device based on information entropy, terminal and storage medium |
| CN118329196A (en)* | 2024-06-17 | 2024-07-12 | 济南红太阳数控设备有限公司 | Cutter vibration signal detection method of vibrating cutter cutting machine |
| Publication | Publication Date | Title |
|---|---|---|
| CN110000610A (en) | A kind of Tool Wear Monitoring method based on multi-sensor information fusion and depth confidence network | |
| CN108356606B (en) | On-line monitoring method of tool wear based on wavelet packet analysis and RBF neural network | |
| CN110647943B (en) | A cutting tool wear monitoring method based on evolutionary data cluster analysis | |
| CN111300146A (en) | Numerical control machine tool cutter abrasion loss online prediction method based on spindle current and vibration signal | |
| CN114619292B (en) | A Milling Tool Wear Monitoring Method Based on Wavelet Noise Reduction and Attention Mechanism Fusion GRU Network | |
| CN109514349B (en) | Tool wear state monitoring method based on vibration signal and Stacking integrated model | |
| CN108319962B (en) | Tool wear monitoring method based on convolutional neural network | |
| CN109434564A (en) | A kind of cutter wear state monitoring method based on deep neural network | |
| CN104723171B (en) | Cutter wear monitoring method based on current and acoustic emission compound signals | |
| CN103105820A (en) | Machining cutter abrasion state identification method of numerical control machine tool | |
| CN106002490B (en) | Milling workpiece roughness monitoring method based on Path and redundant eliminating | |
| CN108073158A (en) | Based on PCA and KNN density algorithm Wind turbines Method for Bearing Fault Diagnosis | |
| CN105834834B (en) | Cutter wear state monitoring method based on drosophila optimized algorithm | |
| CN110346130B (en) | Boring flutter detection method based on empirical mode decomposition and time-frequency multi-feature | |
| WO2016004749A1 (en) | Method for recognizing tool abrasion degree of large numerical control milling machine | |
| CN113485244A (en) | Numerical control machine tool control system and method based on cutter wear prediction | |
| CN114346762B (en) | Method, system and storage medium for chatter monitoring during high-speed milling | |
| CN116372665A (en) | A tool health monitoring method for aviation CNC machine tools based on OPC UA | |
| CN111331429A (en) | Method and device for monitoring tool wear state based on wavelet packet energy analysis | |
| Han et al. | An intelligent fault diagnosis method of variable condition gearbox based on improved DBN combined with WPEE and MPE | |
| CN116871978A (en) | Drilling tool condition monitoring method based on multi-sensor fusion | |
| CN109597315A (en) | A kind of mechanical equipment health degenerate state discrimination method, equipment and system | |
| CN113043073A (en) | Cutter abrasion and service life prediction method and device | |
| CN113369993B (en) | Method for monitoring tool wear state under small sample | |
| CN114871850B (en) | A Tool Wear State Evaluation Method Based on Vibration Signal and BP Neural Network |
| 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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | Application publication date:20190712 |