WO2022268043A1 - 一种基于混合神经模型的数控机床刀具剩余寿命预测方法 - Google Patents
一种基于混合神经模型的数控机床刀具剩余寿命预测方法 Download PDFInfo
- Publication number
- WO2022268043A1 WO2022268043A1 PCT/CN2022/099990 CN2022099990W WO2022268043A1 WO 2022268043 A1 WO2022268043 A1 WO 2022268043A1 CN 2022099990 W CN2022099990 W CN 2022099990W WO 2022268043 A1 WO2022268043 A1 WO 2022268043A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- tool
- layer
- neural network
- life
- data
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000001537 neural effect Effects 0.000 title claims abstract description 16
- 239000013598 vector Substances 0.000 claims abstract description 45
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 24
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000005070 sampling Methods 0.000 claims abstract description 16
- 238000003062 neural network model Methods 0.000 claims abstract description 10
- 230000006403 short-term memory Effects 0.000 claims abstract description 9
- 230000004927 fusion Effects 0.000 claims abstract description 4
- 239000010410 layer Substances 0.000 claims description 87
- 238000011176 pooling Methods 0.000 claims description 26
- 230000008569 process Effects 0.000 claims description 17
- 238000012549 training Methods 0.000 claims description 17
- 230000007774 longterm Effects 0.000 claims description 9
- 238000012795 verification Methods 0.000 claims description 8
- 230000002159 abnormal effect Effects 0.000 claims description 7
- 230000007423 decrease Effects 0.000 claims description 7
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 210000002569 neuron Anatomy 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 230000015654 memory Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 239000002356 single layer Substances 0.000 claims description 4
- 230000009471 action Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims 1
- 230000003993 interaction Effects 0.000 claims 1
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 238000005520 cutting process Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011065 in-situ storage Methods 0.000 description 1
- 238000003754 machining Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000002285 radioactive effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/18—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form
- G05B19/406—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form characterised by monitoring or safety
- G05B19/4065—Monitoring tool breakage, life or condition
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23Q—DETAILS, COMPONENTS, OR ACCESSORIES FOR MACHINE TOOLS, e.g. ARRANGEMENTS FOR COPYING OR CONTROLLING; MACHINE TOOLS IN GENERAL CHARACTERISED BY THE CONSTRUCTION OF PARTICULAR DETAILS OR COMPONENTS; COMBINATIONS OR ASSOCIATIONS OF METAL-WORKING MACHINES, NOT DIRECTED TO A PARTICULAR RESULT
- B23Q17/00—Arrangements for observing, indicating or measuring on machine tools
- B23Q17/09—Arrangements for observing, indicating or measuring on machine tools for indicating or measuring cutting pressure or for determining cutting-tool condition, e.g. cutting ability, load on tool
- B23Q17/0995—Tool life management
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/32—Operator till task planning
- G05B2219/32335—Use of ann, neural network
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/37—Measurements
- G05B2219/37252—Life of tool, service life, decay, wear estimation
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/37—Measurements
- G05B2219/37434—Measuring vibration of machine or workpiece or tool
Definitions
- the invention belongs to the field of life prediction of CNC machine tools, in particular to a method for predicting the remaining life of CNC machine tools based on a mixed neural model.
- Tool wear measurement methods can be roughly divided into direct method and indirect method.
- Direct measurement requires measuring the actual wear, using different methods such as: optical measurement, radioactive analysis and electrical resistance measurement.
- directly measuring tool wear between or during machining operations is difficult.
- the other is indirect measurement.
- the indirect method of detecting the state of the tool based on the relationship between the condition of the tool and the measurable signal has been extensively studied. For example, by using force, vibration and acoustic emission (AE) signals, Sun et al. predicted the remaining life value of cutting tools based on operational reliability assessment and back-propagation neural network (BPNN).
- AE force, vibration and acoustic emission
- BPNN back-propagation neural network
- Zhou et al. proposed a method for predicting the remaining service life of tools under variable working conditions based on LSTM network. According to the factors affecting tool wear, a unified representation of working conditions is established, wear characteristics are extracted from process monitoring signals, and feature vectors are constructed by using the extracted wear characteristics and other working conditions to capture complex spatio-temporal relationships under variable working conditions.
- Zhao et al. proposed a local feature-based gated regression unit (LFGRU) network, which further designed an improved bidirectional GRU to automatically extract deep features after manual features, and added a supervised learning layer to the last layer of the neural network to predict the machine state.
- LFGRU local feature-based gated regression unit
- CLSTM convolution-based long-term short-term memory
- CNN convolutional neural network
- LSTM LSTM
- the proposed network performs convolution operation on the input-to-state and state-to-steady-state transition of LSTM, which contains the time-frequency and time information of the signal, not only retains the advantages of LSTM, but also contains time-frequency characteristics.
- the convolutional structure in LSTM has the ability to capture long-term dependencies while extracting features from the time-frequency domain.
- the present invention proposes a method for predicting the remaining life of a CNC machine tool based on a hybrid neural model, including constructing a hybrid neural network model, which includes a convolutional neural network, a long-term short-term memory network, and an NFM neural network, and performs a numerical control machine tool Tool remaining life prediction specifically includes the following steps:
- S1 Aiming at the tool data sampling frequency, construct the samples of PLC working condition signal data and vibration and current signals in the same time period, and generate a sample data based on the low-frequency PLC sampling points;
- S2 Use the combined sub-model of the convolutional neural network and the long-term short-term memory network to learn the sample data to obtain the first feature vector representing the tool life;
- the process of learning and obtaining the first eigenvector representing tool life includes:
- S21 First, use the triangle window to obtain the mean value of the vibration data collected by the sensor, extract the variance to detect abnormal values in the rectangle window, fill the abnormal points with the mean value, perform sliding average sequence filtering on the collected current signal, and standardize the overall data;
- S22 Build a convolutional neural network layer, and perform convolution operations on the input current signal and vibration signal sequences through K convolution kernels in the convolution layer to obtain K feature vectors, and use the maximum pooling layer to perform convolution operations on the inputted current signal and vibration signal sequences.
- Each feature vector of is pooled, and a time series containing K local features is obtained;
- S24 Use the Global Average Pooling layer and the Global Max Pooling layer to obtain N eigenvalues of LSTM neurons respectively, and use the 2*N eigenvalues as the first eigenvectors representing tool life.
- the convolutional neural network layer includes 3 convolutional units, and one convolutional unit includes a convolutional layer and a maximum pooling layer, and a total of three convolutional units are used.
- the process of obtaining the second eigenvector representing the tool life includes:
- I is the index number
- M is the embedding dimension
- x i is the input vector
- v i is the mapping vector
- I is the index number
- M is the embedding dimension
- x i is the input vector
- v i is the mapping vector
- n is the index number
- ⁇ represents the element product of two vectors.
- the Dropout layer is added after the pooling layer of the convolutional neural network, the Bi-Interaction pooling layer of the NFM neural network, and each fully connected layer, and the LSTM layer and the full connection layer are added.
- the method for predicting the remaining life of a CNC machine tool tool based on a hybrid neural model proposed by the invention can effectively and accurately predict tool life according to the sparse features in the tool.
- Fig. 1 is a schematic flow chart of the method for predicting the remaining life of a CNC machine tool according to the present invention
- Fig. 2 is a frame diagram of the model described in the embodiment of the present invention.
- Fig. 3 is a graph showing the prediction curve of the service life value of the tool 1 in the embodiment of the present invention.
- Fig. 4 is a graph showing the prediction curve of the service life value of the tool 2 in the embodiment of the present invention.
- Fig. 5 is a graph showing the prediction curve of the service life value of the tool 3 in the embodiment of the present invention.
- the present invention proposes a method for predicting the remaining life of a CNC machine tool tool based on a hybrid neural model, and constructs a hybrid neural network model, which includes a convolutional neural network, a long-term short-term memory network, and an NFM neural network, and performs prediction of the remaining life of a CNC machine tool tool specifically including The following steps:
- S1 Aiming at the tool data sampling frequency, construct the samples of PLC working condition signal data and vibration and current signals in the same time period, and generate a sample data based on the low-frequency PLC sampling points;
- S2 Use the combined sub-model of the convolutional neural network and the long-term short-term memory network to learn the sample data to obtain the first feature vector representing the tool life;
- the process of learning and obtaining the first eigenvector representing tool life includes:
- S21 First, use the triangle window to obtain the mean value of the vibration data collected by the sensor, extract the variance to detect abnormal values in the rectangle window, fill the abnormal points with the mean value, perform sliding average sequence filtering on the collected current signal, and standardize the overall data;
- S22 Build a convolutional neural network layer, and perform convolution operations on the input current signal and vibration signal sequences through K convolution kernels in the convolution layer to obtain K feature vectors, and use the maximum pooling layer to perform convolution operations on the inputted current signal and vibration signal sequences.
- Each feature vector of is pooled, and a time series containing K local features is obtained;
- S24 Use the Global Average Pooling layer and the Global Max Pooling layer to obtain N eigenvalues of LSTM neurons respectively, and use the 2*N eigenvalues as the first eigenvectors representing tool life.
- the convolutional neural network layer includes 3 convolutional units, and one convolutional unit includes a convolutional layer and a maximum pooling layer, and a total of three convolutional units are used.
- the process of obtaining the second eigenvector representing the tool life includes:
- I is the index number
- M is the embedding dimension
- x i is the input vector
- v i is the mapping vector
- I is the index number
- M is the embedding dimension
- x i is the input vector
- v i is the mapping vector
- n is the index number
- ⁇ represents the element product of two vectors.
- the Dropout layer is added after the pooling layer of the convolutional neural network, the Bi-Interaction pooling layer of the NFM neural network, and each fully connected layer, and the LSTM layer and the full connection layer are added.
- the original signal data is preprocessed, that is, the vibration signal is subjected to a sliding window to remove abnormal values and the current signal is subjected to sliding filtering, and then the data is normalized; similarly, the data of the working condition signal collected by the PLC is also to normalize.
- the high-frequency data and low-frequency data are combined together according to the low sampling frequency, and every 776 sampling points of high-frequency data are combined with 1 low-frequency sampling point to generate a sample.
- the high-frequency part of the neural network model uses the structure of convolutional neural network (CNN) and long-term short-term memory (LSTM) for feature extraction.
- CNN convolutional neural network
- LSTM long-term short-term memory
- the three convolutional layers and the maximum pooling layer of CNN are used to obtain deep temporal features from the original data.
- the number of convolution kernels of the three convolutional layers is set to 32, 64, and 128 respectively, and each convolutional layer is connected with A dropout layer is added between the pooling layers, and its parameters are set to 0.3, 0.35, and 0.45 respectively, and then a single-layer LSTM is used to mine features after the convolutional layer.
- the number of neurons in the LSTM is set to 128, and after the LSTM Add the Batch Normalization layer and the dropout layer, and set the parameters of the dropout layer to 0.45.
- the LSTM layer use the GlobalAveragePooling1D layer and the GlobalMaxPooling1D layer to obtain the Max Pooling pool vector and the Mean Pooling pool vector respectively; for low-frequency data, use the neural factorization machine (NFM) structure
- NFM neural factorization machine
- the input PLC working condition signal is hash-encoded to obtain a vector table representing the index, where the dimension of the vector table is set to 32, and then the Bi-Interaction Pooling pool vector is obtained by using the BiInteractionPooling layer of NFM.
- the obtained high-frequency information features and low-frequency information features are sent to the three-layer neural network, and the hidden layer neurons are set to 128 and 64 respectively.
- a dropout layer is also added.
- the parameter of the dropout layer is set to 0.45, and finally forms a complete neural network structure with the percentage of remaining life value.
- the training data is divided into a training data set and a verification data set, and the average absolute value error loss function is selected as the training error in the training, and Adam is used for the optimization method, and the initial learning rate is set to 0.0001, adjust the learning rate to the original 0.1 every 5 epochs when the learning rate does not decrease, and use the early stop method to get the optimal model.
- the number of early stop steps is set to 9 steps, that is, when the loss of the verification set is at Stop training when the error no longer decreases in 9 training sessions.
- the tool remaining life prediction experiment was carried out by analyzing the data of different collection time intervals of three tools under different working conditions, among which tool 1 only provided the collected data from the 40th minute to the 90th minute of working time, and tool 2 only provided the working time For data collection from the 70th minute to the 120th minute, the tool 3 provides a working interval from the 50th minute to the 100th minute.
- the trends of the remaining life prediction results of the three test tools are shown in Figure 2, the abscissa is the current working time, and the ordinate is the corresponding remaining life value.
- Figures 3 to 5 respectively provide the actual service life of cutter 1, cutter 2 and cutter 3 and the prediction situation of cutter service life by the present invention (NFM-CLSTM), CNN and CNN-LSTM models respectively, according to the curve in the figure can be It can be seen that the prediction method of the present invention is closer to the actual value for the prediction of the tool life than the CNN and CNN-LSTM models.
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Manufacturing & Machinery (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Mechanical Engineering (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Feedback Control In General (AREA)
Abstract
本发明属于数控机床刀具寿命预测领域,特别涉及一种基于混合神经模型的数控机床刀具剩余寿命预测方法,包括构建混合神经网络模型,针对刀具数据采样频率,构建PLC工况信号数据与振动和电流信号的同一时间段的样本,生成一个样本数据;利用卷积神经网络和长短期记忆网络联合的子模型对样本数据进行学习得到第一表征刀具寿命的特征向量;利用NFM神经网络将采样点的工况信号hash成一独特的索引值后生成一特定维度的可学习的向量表,学习获得第二表征刀具寿命的特征向量;将刀具当前工作时长和获取的特征向量输入到多层感知机中进行融合,预测刀具寿命;本发明能够根据刀具中的稀疏特征对刀具寿命进行有效、精确的预测。
Description
本申请要求于2021年6月22日提交中国专利局、申请号为2021106906541、发明名称为“一种基于混合神经模型的数控机床刀具剩余寿命预测方法”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本发明属于数控机床刀具寿命预测领域,特别涉及一种基于混合神经模型的数控机床刀具剩余寿命预测方法。
刀具作为在工业制造过程中的重要工具,其寿命和磨损状态影响着工件的生产质量,生产效率以及车床的健康状态。如果能精准预测出刀具的剩余寿命,将有效地降低工业制造的成本。刀具磨损测量方法大致可分为直接法和间接法两种。直接测量需要测量实际磨损,使用不同的方法,如:光学测量,放射性分析和电阻测量。然而,直接测量加工操作之间或期间的刀具磨损是困难的。另外就是间接测量,根据刀具条件和可测量信号(如力、声发射、振动、电流等)之间的关系来检测刀具状态的间接方法已经得到了广泛的研究。例如,通过使用力,振动和声发射(AE)信号,Sun等人基于运行可靠性评估和反向传播神经网络(BPNN)预测了切削刀具的剩余寿命值。
传统机器学习方法需要手工提取特征,有很大的局限性,而深度学习模型能够从实时地大量数据中自动抽取特征,并且深层的网络结构赋予其强大的非线性学习能力,有效弥补了传统机器学习方法的缺陷。Zhou等提出了一种基于LSTM网络的可变工况下刀具剩余使用寿命预测方法。根据影响刀具磨损的因素,建立工作条件的统一表示,从过程监测信号中提取磨损特性,利用提取的磨损特征等工况构造成特征向量,捕捉可变工况下复杂的时空关系。基于LSTM模型在解决具有复杂相关性和记忆积累效应的问题中的独特优势,建立了可变工 况下的工具剩余使用寿命预测模型。Zhao等提出了基于局部特征的门控回归单元(LFGRU)网络,其在手工特征后进一步设计了改进的双向GRU自动化提取深层特征,在神经网络末层加入一个监督学习层来预测机器状态。通过刀具磨损预测、齿轮箱故障诊断和初期轴承故障检测三个机器健康监测任务的实验,验证了所提出的LFGRU的有效性和泛化能力。Meng等提出了一种新的基于卷积的长期短时记忆(CLSTM)网络来预测旋转机械挖掘原位振动数据的RUL,与简单地将卷积神经网络(CNN)串行连接到长短时记忆(LSTM)网络的研究不同,所提出的网络对LSTM的输入到状态和状态到稳态转换进行卷积运算,该转换包含信号的时频和时间信息,不仅保留了LSTM的优点,而且还包含了时频特征。在LSTM中的卷积结构具有捕获长期依赖关系的能力,同时还能从时频域提取特征。通过逐层叠加多个CLSTM,形成编码预测体系结构,建立了RUL预测的深度学习模型,Sun等提出了一种基于稀疏自编码器的深度传输学习(DTL)网络。在DTL方法中,采用了三种转移策略,即权重转移、隐藏特征转移学习和权重更新,将由历史故障数据训练的SAE转移到新对象。通过这些策略,实现了对无监督信息的新目标的预测训练。Zhang等提出了一种基于深度学习的动态系统性能跟踪和后续RUL预测方法。利用LSTM作为模型,因其具备在发现时间序列的变化模式方面有着强大的功能,被用来跟踪系统的退化。然而目前研究中并没有针对刀具中的稀疏特征提出很好的解决方法,本专利针对此进行了研究。
发明内容
针对刀具中的稀疏特征,本发明提出基于混合神经模型的数控机床刀具剩余寿命预测方法,包括构建混合神经网络模型,该模型包括卷积神经网络、长短期记忆网络以及NFM神经网络,进行数控机床刀具剩余寿命预测具体包括以下步骤:
S1:针对刀具数据采样频率,构建PLC工况信号数据与振动和电流信号的同一时间段的样本,以低频率PLC采样点为标准生成一个样本数据;
S2:利用卷积神经网络和长短期记忆网络联合的子模型对样本数据进行学习得到第一表征刀具寿命的特征向量;
S3:利用NFM神经网络将每一采样点的工况信号hash成一独特的索引值后生成一特定维度的可学习的向量表,通过Bi-Interaction Pooling层学习获得第二表征刀具寿命的特征向量;
S4:将刀具当前工作时长、第一表征刀具寿命的特征向量和第二表征刀具寿命的特征向量输入到多层感知机中进行融合,然后对整个网络结构进行学习。
进一步的,学习得到第一表征刀具寿命的特征向量的过程包括:
S21:先对传感器采集的振动数据利用triangle窗口获取均值,rectangle窗口提取方差检测异常值,用均值填充异常点,对采集的电流信号进行滑动平均序列滤波,并对整体数据进行标准化;
S22:搭建卷积神经网络层,通过卷积层的K个卷积核对输入的电流信信号和振动信号序列进行卷积操作,得到K个特征向量,通过最大池化层对经卷积层后的每个特征向量进行池化,得到了包含K个局部特征的时间序列;
S23:搭建单层长短期记忆层,通过包含N个LSTM的神经单元的LSTM层,得到了含有历史时间信息的时间序列;
S24:利用Global Average Pooling层和Global Max Pooling层分别获取LSTM神经元的N个特征值,将该2*N个特征值作为第一表征刀具寿命的特征向量。
进一步的,卷积神经网络层包括3个卷积单元,一个卷积单元包括一个卷积层和一个最大池化层,共采用了三个卷积单元。
进一步的,获取第二表征刀具寿命的特征向量的过程包括:
进一步的,在对混合神经网络模型进行训练过程中,在卷积神经网络的池化层、NFM神经网络的Bi-Interaction池化层以及每个全连接层后加入Dropout层,在LSTM层和全连接层后加入Batch Normalization层;在神经网络反向传播中利用Adam来对最优模型进行寻优,在寻优过程中epoch验证数据的loss每n次不再下降时便降低学习率,且在训练时验证数据的loss至少N次不再下降时就停止训练。
本发明提出的基于混合神经模型的数控机床刀具剩余寿命预测方法能够根据刀具中的稀疏特征对道具寿命进行有效、精确的预测。
图1为本发明所述数控机床刀具剩余寿命预测方法流程示意图;
图2为本发明实施例所述模型框架图;
图3为本发明实施例中刀具1使用寿命值预测曲线图;
图4为本发明实施例中刀具2使用寿命值预测曲线图;
图5为本发明实施例中刀具3使用寿命值预测曲线图。
下面将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。
本发明提出一种基于混合神经模型的数控机床刀具剩余寿命预测方法,构建混合神经网络模型,该模型包括卷积神经网络、长短期记忆网络以及NFM神经网络,进行数控机床刀具剩余寿命预测具体包括以下步骤:
S1:针对刀具数据采样频率,构建PLC工况信号数据与振动和电流信号的同一时间段的样本,以低频率PLC采样点为标准生成一个样本数据;
S2:利用卷积神经网络和长短期记忆网络联合的子模型对样本数据进行学习得到第一表征刀具寿命的特征向量;
S3:利用NFM神经网络将每一采样点的工况信号hash成一独特的索引值后生成一特定维度的可学习的向量表,通过Bi-Interaction Pooling层学习获得第二表征刀具寿命的特征向量;
S4:将刀具当前工作时长、第一表征刀具寿命的特征向量和第二表征刀具寿命的特征向量输入到多层感知机中进行融合,然后对整个网络结构进行学习。
进一步的,学习得到第一表征刀具寿命的特征向量的过程包括:
S21:先对传感器采集的振动数据利用triangle窗口获取均值,rectangle窗口提取方差检测异常值,用均值填充异常点,对采集的电流信号进行滑动平均序列滤波,并对整体数据进行标准化;
S22:搭建卷积神经网络层,通过卷积层的K个卷积核对输入的电流信信号和振动信号序列进行卷积操作,得到K个特征向量,通过最大池化层对经卷积层后的每个特征向量进行池化,得到了包含K个局部特征的时间序列;
S23:搭建单层长短期记忆层,通过包含N个LSTM的神经单元的LSTM层,得到了含有历史时间信息的时间序列;
S24:利用Global Average Pooling层和Global Max Pooling层分别获取LSTM神经元的N个特征值,将该2*N个特征值作为第一表征刀具寿命的特征向量。
进一步的,卷积神经网络层包括3个卷积单元,一个卷积单元包括一个卷积层和一个最大池化层,共采用了三个卷积单元。
进一步的,获取第二表征刀具寿命的特征向量的过程包括:
进一步的,在对混合神经网络模型进行训练过程中,在卷积神经网络的池化层、NFM神经网络的Bi-Interaction池化层以及每个全连接层后加入Dropout层,在LSTM层和全连接层后加入Batch Normalization层;在神经网络反向传播中利用Adam来对最优模型进行寻优,在寻优过程中epoch验证数据的loss每n次不再下降时便降低学习率,且在训练时验证数据的loss至少N次不再下降时就停止训练。
实施例1
请参阅图1和图2所示,收集PLC控制器信号和外置传感器信号,监测加工过程中的工况信息和传感器数据,传感器数据主要为电流信号和三个方向即x轴、y轴以及z轴的振动信号,以实现刀具磨损在线监测与寿命预测为目标;数据采集过程从一把全新刀具执行加工程序时开始,直到刀具寿命终止时停止,其中PLC信号的采样频率为33Hz,振动传感器的采样频率25600Hz。数据集每隔5分钟提供了一个1分钟的片段作为样本,由时间序列1.csv,2.csv,…,n.csv给出。
在本实施例中对原始信号数据进行预处理,即对振动信号进行滑动窗口去除异常值和对电流信号进行滑动滤波,然后对数据进行归一化;同样对PLC采集的工况信号的数据也进行归一化。将高频数据和低频数据按照低采样频率合并到一起,对高频数据每776个采样点与低频1个采样点合并生成一个样本。
构建混合神经网络模型,即对刀具信号与寿命之间的关系进行建模,其中神经网络模型高频部分结构,采用卷积神经网络(CNN)与长短期记忆(LSTM)的结构进行特征提取,利用CNN的三个卷积层和最大池化层从原始数据中获取深层次时序特征,三个卷积层的卷积核数量分别设置为32,64,128,且在每个卷积层与池化层之间加入dropout层,其参数分别设置为0.3、0.35、0.45,然后在卷积层后利用单层LSTM对其进行特征挖掘,LSTM的神经元个数设置为128,且在LSTM后加入Batch Normalization层和dropout层,dropout层的参数设置为0.45,在LSTM层后利用GlobalAveragePooling1D层和GlobalMaxPooling1D层分别获取Max Pooling池向量和Mean Pooling池向量;针对低频数据采用神经因子分解机(NFM)结构来进行特征挖掘,将输入的PLC工况信号通过hash编码后,获得一个表征索引的向量表,其中向量表的维度设置为32,然后利用NFM的BiInteractionPooling层获得Bi-Interaction Pooling池向量。
将获得的高频信息特征和低频信息特征送入三层神经网络中,其隐藏层神经分别设置为128、64。每层神经网络后也加入dropout层,dropout层的参数设置为0.45,最后与剩余寿命值百分比构成完整的神经网络结构。
在对混合神经网络模型进行训练过程中,将训练数据划分为训练数据集和验证数据集,在训练中选择平均绝对值误差损失函数作为训练误差,寻优方式才用Adam,初始学习率设置为0.0001,每隔5个epoch学习率不下降时便将学习率调整为原来的0.1,并采用早停的方法来得到最优的模型,早停步数设置为9步即当验证集的loss在9次训练中误差不再下降时便停止训练。
为了验证该方法的可行性和准确性,进行了测试实验,并将此模型与CNN和CNN-LSTM模型进行了对比。通过对三把不同工况下刀具不同采集时间区段 的数据来进行刀具剩余寿命预测实验,其中刀具1只提供了工作时间从第40分钟至第90分钟的采集数据,刀具2只提供了工作时间从第70分钟至第120分钟的采集数据,刀具3提供了第50分钟至第100分钟的工作区间。三把测试刀具的剩余寿命预测结果趋势分别如附图2所示,横坐标为当前工作时间,纵坐标为对应的剩余寿命值。针对预测效果采取指数转换精度(ETA)、RSME和精准度衡量,对比结果如表1。图3~5分别给出刀具1、刀具2以及刀具3的实际使用寿命以及分别通过本发明(NFM-CLSTM)、CNN和CNN-LSTM模型对刀具使用寿命的预测情况,根据图中的曲线可以看出本发明预测方法与CNN和CNN-LSTM模型相比,对刀具使用寿命的预测更加贴近实际值。
表1 不同神经网络预测误差结果对比表
尽管已经示出和描述了本发明的实施例,对于本领域的普通技术人员而言,可以理解在不脱离本发明的原理和精神的情况下可以对这些实施例进行多种变化、修改、替换和变型,本发明的范围由所附权利要求及其等同物限定。
Claims (5)
- 基于混合神经模型的数控机床刀具剩余寿命预测方法,其特征在于,构建混合神经网络模型,该模型包括卷积神经网络、长短期记忆网络以及NFM神经网络,进行数控机床刀具剩余寿命预测具体包括以下步骤:S1:针对刀具数据采样频率,构建PLC工况信号数据与振动和电流信号的同一时间段的样本,以低频率PLC采样点为标准生成一个样本数据;S2:利用卷积神经网络和长短期记忆网络联合的子模型对样本数据进行学习得到第一表征刀具寿命的特征向量,具体包括以下步骤:S21:先对传感器采集的振动数据利用triangle窗口获取均值,rectangle窗口提取方差检测异常值,用均值填充异常点,对采集的电流信号进行滑动平均序列滤波,并对整体数据进行标准化;S22:搭建卷积神经网络层,通过卷积层的K个卷积核对输入的电流信信号和振动信号序列进行卷积操作,得到K个特征向量,通过最大池化层对经卷积层后的每个特征向量进行池化,得到了包含K个局部特征的时间序列;S23:搭建单层长短期记忆层,通过包含N个LSTM的神经单元的LSTM层,得到了含有历史时间信息的时间序列;S24:利用Global Average Pooling层和Global Max Pooling层分别获取LSTM神经元的N个特征值,将该2*N个特征值作为第一表征刀具寿命的特征向量;S3:利用NFM神经网络将每一采样点的工况信号hash成一索引值后生成I*M维度的可学习的向量表,其中I为索引数,M为嵌入维度,通过Bi-Interaction Pooling层学习获得第二表征刀具寿命的特征向量;S4:将刀具当前工作时长、第一表征刀具寿命的特征向量和第二表征刀具寿命的特征向量输入到多层感知机中进行融合,然后对整个网络结构进行学习。
- 根据权利要求1所述的基于混合神经模型的数控机床刀具剩余寿命预测方法,其特征在于,卷积神经网络层包括3个卷积单元,一个卷积单元包括一个卷积层和一个最大池化层。
- 根据权利要求1所述的基于混合神经模型的数控机床刀具剩余寿命预测方法,其特征在于,在对混合神经网络模型进行训练过程中,在卷积神经网络的池化层、NFM神经网络的Bi-Interaction池化层以及每个全连接层后加入Dropout层,在LSTM层和全连接层后加入Batch Normalization层;在神经网络反向传播中利用Adam来对最优模型进行寻优,在寻优过程中epoch验证数据的loss每n次不再下降时便降低学习率,且在训练时验证数据的loss至少N次不再下降时就停止训练。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US18/498,974 US11927937B1 (en) | 2021-06-22 | 2023-10-31 | Prediction method for tool remaining life of numerical control machine tool based on hybrid neural model |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110690654.1A CN113305645B (zh) | 2021-06-22 | 2021-06-22 | 一种基于混合神经模型的数控机床刀具剩余寿命预测方法 |
CN202110690654.1 | 2021-06-22 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US18/498,974 Continuation US11927937B1 (en) | 2021-06-22 | 2023-10-31 | Prediction method for tool remaining life of numerical control machine tool based on hybrid neural model |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022268043A1 true WO2022268043A1 (zh) | 2022-12-29 |
Family
ID=77380056
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2022/099990 WO2022268043A1 (zh) | 2021-06-22 | 2022-06-21 | 一种基于混合神经模型的数控机床刀具剩余寿命预测方法 |
Country Status (3)
Country | Link |
---|---|
US (1) | US11927937B1 (zh) |
CN (1) | CN113305645B (zh) |
WO (1) | WO2022268043A1 (zh) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115922443A (zh) * | 2023-01-09 | 2023-04-07 | 湖南大学 | 一种基于滑动窗口的刀具剩余使用寿命实时预测方法 |
CN116825243A (zh) * | 2023-05-09 | 2023-09-29 | 安徽工程大学 | 一种基于多源数据的热障涂层服役寿命预测方法及系统 |
CN117452236A (zh) * | 2023-12-26 | 2024-01-26 | 深圳市北测检测技术有限公司 | 新能源汽车的电池寿命检测方法及系统 |
CN117961645A (zh) * | 2024-03-29 | 2024-05-03 | 中国矿业大学 | 一种面向小样本噪声数据的刀具磨损软测量方法 |
CN118092194A (zh) * | 2024-04-25 | 2024-05-28 | 中科航迈数控软件(深圳)有限公司 | 机床控制策略的调整方法、控制策略的调整装置及介质 |
CN118348917A (zh) * | 2024-04-15 | 2024-07-16 | 陕西优鹏安盈科技有限责任公司 | 一种基于组态计算机的数控系统实时监测方法 |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113305645B (zh) * | 2021-06-22 | 2022-07-15 | 重庆邮电大学工业互联网研究院 | 一种基于混合神经模型的数控机床刀具剩余寿命预测方法 |
CN113927370A (zh) * | 2021-09-22 | 2022-01-14 | 南京理工大学 | 一种基于多源信号的产线用刀具剩余寿命实时云监测系统及方法 |
CN115446663B (zh) * | 2022-10-14 | 2023-10-20 | 常州先进制造技术研究所 | 基于物理指导的深度学习网络的刀具磨损状态监测方法及应用 |
CN118332279A (zh) * | 2024-04-10 | 2024-07-12 | 大连交通大学 | 基于改进lstmnn的滚动轴承寿命预测 |
CN118445760B (zh) * | 2024-07-08 | 2024-09-13 | 广东工业大学 | 一种新能源汽车的电动机轴承健康状态监测方法及装置 |
CN118656758B (zh) * | 2024-08-16 | 2024-10-25 | 浙江大学 | 基于孪生数据多域挖掘的装备部件状态可信感知方法 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8781982B1 (en) * | 2011-09-23 | 2014-07-15 | Lockheed Martin Corporation | System and method for estimating remaining useful life |
CN110153802A (zh) * | 2019-07-04 | 2019-08-23 | 西南交通大学 | 一种基于卷积神经网络和长短时记忆神经网络联合模型的刀具磨损状态辨识方法 |
CN110472800A (zh) * | 2019-08-23 | 2019-11-19 | 山东浪潮通软信息科技有限公司 | 一种基于lstm+cnn的机床刀具剩余寿命预测方法 |
CN112631128A (zh) * | 2020-11-27 | 2021-04-09 | 广东省科学院智能制造研究所 | 一种多模异构信息融合的机器人装配技能学习方法及系统 |
CN113305645A (zh) * | 2021-06-22 | 2021-08-27 | 重庆邮电大学工业互联网研究院 | 一种基于混合神经模型的数控机床刀具剩余寿命预测方法 |
Family Cites Families (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10761894B2 (en) * | 2017-10-30 | 2020-09-01 | Sas Institute Inc. | Methods and systems for automated monitoring and control of adherence parameters |
CN109465676B (zh) * | 2018-09-13 | 2020-10-16 | 成都数之联科技有限公司 | 一种刀具寿命预测方法 |
CN109522600B (zh) * | 2018-10-16 | 2020-10-16 | 浙江大学 | 基于组合深度神经网络的复杂装备剩余使用寿命预测方法 |
CN109822399B (zh) * | 2019-04-08 | 2020-07-14 | 浙江大学 | 基于并行深度神经网络的数控机床刀具磨损状态预测方法 |
CN110263474A (zh) * | 2019-06-27 | 2019-09-20 | 重庆理工大学 | 一种数控机床的刀具寿命实时预测方法 |
CN112150094A (zh) * | 2019-06-28 | 2020-12-29 | 华为技术有限公司 | 模型训练方法、基于模型的评估方法及装置 |
CN110303380B (zh) * | 2019-07-05 | 2021-04-16 | 重庆邮电大学 | 一种数控机床刀具剩余寿命预测方法 |
CA3053894A1 (en) * | 2019-07-19 | 2021-01-19 | Inspectorio Inc. | Defect prediction using historical inspection data |
CN111325112B (zh) * | 2020-01-31 | 2023-04-07 | 贵州大学 | 基于深度门控循环单元神经网络的刀具磨损状态监测方法 |
CN111475921A (zh) * | 2020-03-13 | 2020-07-31 | 重庆邮电大学 | 一种基于边缘计算和lstm网络的刀具剩余寿命预测方法 |
CN111931905B (zh) * | 2020-07-13 | 2024-05-14 | 江苏大学 | 一种图卷积神经网络模型、及利用该模型的车辆轨迹预测方法 |
-
2021
- 2021-06-22 CN CN202110690654.1A patent/CN113305645B/zh active Active
-
2022
- 2022-06-21 WO PCT/CN2022/099990 patent/WO2022268043A1/zh active Application Filing
-
2023
- 2023-10-31 US US18/498,974 patent/US11927937B1/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8781982B1 (en) * | 2011-09-23 | 2014-07-15 | Lockheed Martin Corporation | System and method for estimating remaining useful life |
CN110153802A (zh) * | 2019-07-04 | 2019-08-23 | 西南交通大学 | 一种基于卷积神经网络和长短时记忆神经网络联合模型的刀具磨损状态辨识方法 |
CN110472800A (zh) * | 2019-08-23 | 2019-11-19 | 山东浪潮通软信息科技有限公司 | 一种基于lstm+cnn的机床刀具剩余寿命预测方法 |
CN112631128A (zh) * | 2020-11-27 | 2021-04-09 | 广东省科学院智能制造研究所 | 一种多模异构信息融合的机器人装配技能学习方法及系统 |
CN113305645A (zh) * | 2021-06-22 | 2021-08-27 | 重庆邮电大学工业互联网研究院 | 一种基于混合神经模型的数控机床刀具剩余寿命预测方法 |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115922443A (zh) * | 2023-01-09 | 2023-04-07 | 湖南大学 | 一种基于滑动窗口的刀具剩余使用寿命实时预测方法 |
CN116825243A (zh) * | 2023-05-09 | 2023-09-29 | 安徽工程大学 | 一种基于多源数据的热障涂层服役寿命预测方法及系统 |
CN116825243B (zh) * | 2023-05-09 | 2024-01-16 | 安徽工程大学 | 一种基于多源数据的热障涂层服役寿命预测方法及系统 |
CN117452236A (zh) * | 2023-12-26 | 2024-01-26 | 深圳市北测检测技术有限公司 | 新能源汽车的电池寿命检测方法及系统 |
CN117452236B (zh) * | 2023-12-26 | 2024-03-15 | 深圳市北测检测技术有限公司 | 新能源汽车的电池寿命检测方法及系统 |
CN117961645A (zh) * | 2024-03-29 | 2024-05-03 | 中国矿业大学 | 一种面向小样本噪声数据的刀具磨损软测量方法 |
CN117961645B (zh) * | 2024-03-29 | 2024-05-28 | 中国矿业大学 | 一种面向小样本噪声数据的刀具磨损软测量方法 |
CN118348917A (zh) * | 2024-04-15 | 2024-07-16 | 陕西优鹏安盈科技有限责任公司 | 一种基于组态计算机的数控系统实时监测方法 |
CN118092194A (zh) * | 2024-04-25 | 2024-05-28 | 中科航迈数控软件(深圳)有限公司 | 机床控制策略的调整方法、控制策略的调整装置及介质 |
Also Published As
Publication number | Publication date |
---|---|
US20240061396A1 (en) | 2024-02-22 |
CN113305645B (zh) | 2022-07-15 |
CN113305645A (zh) | 2021-08-27 |
US11927937B1 (en) | 2024-03-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022268043A1 (zh) | 一种基于混合神经模型的数控机床刀具剩余寿命预测方法 | |
CN111596604B (zh) | 一种基于数字孪生的工程装备故障智能诊断与自愈控制系统及方法 | |
Xu et al. | Deep learning-based tool wear prediction and its application for machining process using multi-scale feature fusion and channel attention mechanism | |
Xu et al. | A novel multi-scale CNN and attention mechanism method with multi-sensor signal for remaining useful life prediction | |
Lindemann et al. | Anomaly detection in discrete manufacturing using self-learning approaches | |
CN113255848B (zh) | 基于大数据学习的水轮机空化声信号辨识方法 | |
CN111178553A (zh) | 一种基于arima和lstm算法的工业设备健康趋势分析方法及系统 | |
CN113128561A (zh) | 一种机床轴承故障诊断方法 | |
CN113657221B (zh) | 一种基于智能感知技术的电厂设备状态监测方法 | |
CN109407518A (zh) | 家庭服务机器人运行状态自主认知方法及系统 | |
CN112328703B (zh) | 一种基于增量学习的数控设备健康状态诊断装置及方法 | |
CN113671917A (zh) | 针对多模态工业过程异常状态的检测方法、系统、设备 | |
CN110570013B (zh) | 一种单站位在线波周期数据的预测诊断方法 | |
CN118154174A (zh) | 一种工业设备智慧运维云平台 | |
CN111126255A (zh) | 基于深度学习回归算法的数控机床刀具磨损值预测方法 | |
Xu et al. | Anomaly detection with gru based bi-autoencoder for industrial multimode process | |
Maqbool et al. | An efficient fault-prediction mechanism for improving yield in industry 5.0 | |
CN116821828A (zh) | 一种基于工业数据的多维时序预测方法 | |
Karagiorgou et al. | Unveiling trends and predictions in digital factories | |
Huang et al. | Tool remaining useful life prediction based on edge data processing and lstm recurrent neural network | |
CN116933003A (zh) | 基于DaNet的无人机发动机的剩余使用寿命预测方法 | |
CN114760194A (zh) | 一种物联网设备的故障研判方法 | |
CN114818813A (zh) | 一种基于自注意力卷积神经网络的刀具磨损值预测算法 | |
Zhai et al. | A novel approach based on spatio-temporal attention and multi-scale modeling for mechanical failure prediction | |
Chen et al. | Machine Learning–based Fault Prediction and Diagnosis of Brushless Motors |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22827536 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 22827536 Country of ref document: EP Kind code of ref document: A1 |