WO2024045555A1 - 一种基于标准自学习数据增强的故障诊断方法及系统 - Google Patents

一种基于标准自学习数据增强的故障诊断方法及系统 Download PDF

Info

Publication number
WO2024045555A1
WO2024045555A1 PCT/CN2023/081715 CN2023081715W WO2024045555A1 WO 2024045555 A1 WO2024045555 A1 WO 2024045555A1 CN 2023081715 W CN2023081715 W CN 2023081715W WO 2024045555 A1 WO2024045555 A1 WO 2024045555A1
Authority
WO
WIPO (PCT)
Prior art keywords
fault diagnosis
model
data enhancement
learning
standard self
Prior art date
Application number
PCT/CN2023/081715
Other languages
English (en)
French (fr)
Inventor
安增辉
张玉玺
杨蕊
王后亮
闫英珑
Original Assignee
山东建筑大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 山东建筑大学 filed Critical 山东建筑大学
Publication of WO2024045555A1 publication Critical patent/WO2024045555A1/zh

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Definitions

  • the invention belongs to the technical field of bearing fault diagnosis, and in particular relates to a fault diagnosis method and system based on standard self-learning data enhancement.
  • rolling bearings are the core of internal motion conversion and power transmission in high-end equipment.
  • Rolling bearings often operate under strong non-stationary working conditions.
  • the load and speed fluctuate violently, which on the one hand leads to frequent rolling bearing failures and on the other hand
  • the deepening of the structure means that it is easier to extract accurate fault features, and it also causes the model to overfit the training data, which highlights the importance of complete training data; for incomplete health monitoring data, it is necessary to extract target features.
  • the deepening of the model structure will only make it fall into overfitting with limited diagnostic knowledge and fail to meet actual diagnostic needs. Therefore, complete health monitoring training data is the basic prerequisite for the implementation of intelligent fault diagnosis methods.
  • a complete training data set under strongly non-stationary working conditions requires the superposition completeness of three-dimensional continuous information on faults, instantaneous working conditions (speed, load, etc.), and working condition change rates (speed, load change rate, etc.), that is, each fault Samples must be collected under any instantaneous working conditions and any working condition change rate. Such demanding requirements cannot be achieved in practice. In practice, once a fault is discovered in the equipment, it must be shut down for maintenance to prevent serious accidents. The fault sample is only a period of time.
  • the working condition change rate information is extremely simple, and a certain range of instantaneous working condition information (such as rotation speed) is inevitably missing, which is far from meeting the requirements of completeness; it can be seen that the collection under strongly non-stationary working conditions The training data is extremely incomplete, which seriously restricts the development of intelligent fault diagnosis.
  • Data Augmentation is the most direct method to deal with incomplete data sets by generating new training samples; traditional methods originated from image recognition pre-processing, such as image rotation, amplification, etc.; in recent years, generative adversarial Neural Networks (Generative Adversarial Networks, GAN), as an intelligent data generation method, have become a hot spot for data enhancement; some GAN-based data enhancement methods have also been proposed in the field of intelligent fault diagnosis of rotating machinery; Zhou et al. designed a GAN generator and The discriminator uses a global optimization scheme to generate more samples to deal with the data imbalance problem; Shao et al. and Guo et al. respectively developed a GAN-based auxiliary classifier framework and a multi-label one-dimensional GAN to learn from mechanical sensor signals. And generate data that is closer to reality to solve the problem of insufficient data.
  • GAN Generic Adversarial Networks
  • Existing data enhancement methods mainly address problems such as imbalanced data sets and small data volumes. They expand the data volume by generating samples that are closer to the original data, thereby improving the diagnostic accuracy of the model.
  • data generation for the purpose of data similarity can only result in convergent data; usually, The health monitoring data of rotating machinery when operating under strongly non-stationary operating conditions is only a limited uniform deceleration data set with missing information.
  • Blindly pursuing the similarity of generated data can only expand the amount of data, but cannot make up for the lack of information in the data set; only generating Only with diverse samples can the data set under strongly non-stationary working conditions satisfy the superposition completeness of three-dimensional continuous information. Therefore, the focus of data generation is the difference between the generated data and the original data.
  • the present invention provides a fault diagnosis method and system based on standard self-learning data enhancement, using a one-dimensional convolutional neural network as the basic framework, using incomplete training data sets, and through standard self-learning data.
  • the cross-adversarial training method of learning and data enhancement generates disturbance data, obtains a fault diagnosis model under strongly non-stationary operating conditions, and improves the accuracy of fault diagnosis.
  • one or more embodiments of the present invention provide the following technical solutions:
  • the first aspect of the present invention provides a fault diagnosis method based on standard self-learning data enhancement
  • a fault diagnosis method based on standard self-learning data enhancement including:
  • the fault diagnosis model is trained to obtain a complete data set and an intelligent fault diagnosis model under strongly non-stationary operating conditions
  • the collected vibration signals to be diagnosed are input into the trained intelligent fault diagnosis model to obtain the bearing fault diagnosis results.
  • the one-dimensional convolutional neural network includes multiple convolutional layers, pooling layers and fully connected layers;
  • the convolution layer uses ReLU (Rectified Linear Unit) as the activation function, and the stride of the convolution operation is 1;
  • All convolutional layers are connected to a pooling layer to reduce the dimensionality of the output features of the convolutional layer;
  • the features of the input sample after multi-layer convolution and pooling are flattened into a one-dimensional vector, and then fault diagnosis is performed through three layers of full connection.
  • the standard self-learning aims at learning classification knowledge through repeated input. Use the updated samples to optimize the parameters in the fault diagnosis model, and self-study the evaluation criteria to determine whether the sample is a disturbance sample.
  • the data enhancement is guided by the output of the model itself and generates disturbance samples through sample parameterization and model digitization methods;
  • the criterion for judging a perturbed sample is whether it can interfere with model judgment, specifically: after the sample is input to the model, it can cause a disturbance in the posterior probability of the model.
  • sample parameterization is to regard the samples as model parameters, train the parameters that reduce the objective function through the stochastic gradient descent method, and then export the parameters into the generated samples.
  • the digitization of the model means that the parameters of the fault diagnosis model are regarded as data, and the parameter values are fixed during the training process.
  • the output of the intelligent fault diagnosis model is the posterior probability that the vibration signal to be diagnosed belongs to each fault type.
  • the probabilities are sorted, and the fault type with the highest probability is the final bearing fault diagnosis result.
  • a second aspect of the present invention provides a fault diagnosis system based on standard self-learning data enhancement.
  • a fault diagnosis system based on standard self-learning data enhancement including a model building module, a model training module and a fault diagnosis module;
  • the model building module is configured to: build a fault diagnosis model based on a one-dimensional convolutional neural network
  • the model training module is configured to: train the fault diagnosis model through the cross-adversarial training method of standard self-learning and data enhancement to obtain a complete data set and an intelligent fault diagnosis model under strongly non-stationary operating conditions;
  • the fault diagnosis module is configured to input the collected vibration signals to be diagnosed into the trained intelligent fault diagnosis model to obtain bearing fault diagnosis results.
  • the third aspect of the present invention provides a computer-readable storage medium on which a program is stored.
  • the program is executed by a processor, the fault diagnosis method based on standard self-learning data enhancement as described in the first aspect of the present invention is implemented. step.
  • a fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor.
  • the processor executes the program, a method as described in the first aspect of the present invention is implemented. Steps in a fault diagnosis method based on standard self-learning data enhancement.
  • This invention proposes a standard self-learning data enhancement method, which uses the fault diagnosis model's own prediction results as the evaluation standard for data generation, generates disturbance samples through sample parameterization and model digitization, and expands the data set to make it closer to a complete data set. .
  • This invention uses a one-dimensional convolutional neural network as the basic framework, utilizes incomplete training data sets, and obtains a fault diagnosis model under strongly non-stationary working conditions through a cross-confrontation training method of standard self-learning and data enhancement, thereby improving the efficiency of fault diagnosis. Accuracy.
  • Figure 1(a)-(b) shows examples of humans trained with incomplete data sets identifying regular targets and perturbing samples.
  • Figure 2 is a method flow chart of the first embodiment.
  • Figure 3 is the architecture diagram of the standard self-learning data enhancement method.
  • Figure 4 is the training flow chart of the standard self-learning data enhancement method.
  • Figure 5 shows the fault test bench and fault bearing.
  • Figure 6 shows the rotational speed changes of samples with different health conditions in the TDR data set.
  • Figure 7 shows the diagnostic results of different test sets.
  • Figure 8 is a system structure diagram of the second embodiment.
  • the fault data set collected under strongly non-stationary working conditions is a typical incomplete data set, but the task of the model is often to diagnose under complex and changeable working conditions, which means that most of the test samples are disturbance samples; therefore, the data
  • the purpose of enhancement is to generate perturbation samples to expand the training set, thereby enhancing the completeness of the training set;
  • the purpose of the GAN-based method is to generate regular samples that are similar to the training samples, which is obviously the same as the purpose of data enhancement under strongly non-stationary conditions. Contradictory; therefore, there is an urgent need to change conventional intelligent data enhancement ideas and then propose intelligent data enhancement methods for perturbed data.
  • the first issue is to clarify the criteria for perturbed samples, that is, how to evaluate the generated samples as perturbed samples. It can be seen from the above examples that "flying fish" as a perturbation sample can make the human brain's prediction results erroneous.
  • the present invention proposes a data enhancement method that uses the fault diagnosis model's own prediction results as the evaluation standard for data generation, and generates samples through sample parameterization and model digitization; since the fault diagnosis model's own prediction results are The model is learned through training data, so it is called the standard self-learning data augmentation method SSDA (Standard Self-learned Data Augmentation).
  • SSDA Standard Self-learned Data Augmentation
  • This embodiment discloses a fault diagnosis method based on standard self-learning data enhancement, as shown in Figure 2, which specifically includes:
  • Step S1 Build a fault diagnosis model based on the one-dimensional convolutional neural network
  • the basic model architecture of SSDA adopts the currently widely used One-Dimensional Convolutional Neural Networks (1-D-CNN);
  • 1-D-CNN includes multi-layer convolution layers, pooling layers and fully connected layer, the parameter set of each layer is shown in Table 1:
  • N is the sample dimension. In this study, N is set to 1200 dimensions.
  • the subscript [ ⁇ ] represents the sequence number of the elements in the matrix, and the stride of the convolution operation is 1. Therefore, The dimension is (N l-1 -K l +1) ⁇ M l .
  • S is the pooling length.
  • w l and b l are the weight matrix and bias vector of the fully connected layer respectively;
  • the activation functions of the first two fully connected layers are ReLU activation functions, and the features of the last fully connected layer are passed through the Softmax activation function to obtain the output of the model.
  • C represents the number of fault types), that is, the elements in the output o can be calculated by the following formula:
  • the output o of the model represents the posterior probability that the sample belongs to each fault type. Therefore, the fault type of the sample can be judged based on the model output.
  • the process of converting the sample x into the feature u 1 after being input into the model is abstracted into the mapping ⁇ f , feature
  • Step S2 Train the fault diagnosis model through the cross-adversarial training method of standard self-learning and data enhancement to obtain a complete data set and an intelligent fault diagnosis model under strongly non-stationary operating conditions;
  • SSDA includes two training steps: standard self-learning and data enhancement; in standard self-learning, with the goal of learning classification knowledge, the parameters in the 1-D-CNN model are optimized by repeatedly inputting updated samples. This process is equivalent to the model I taught myself the evaluation criteria for judging whether a sample is a perturbation sample; in data enhancement, through the method of sample parameterization and model digitization, the posterior probability of the model output result is disturbed, thereby generating diversified samples; after two training By alternating the steps, not only can a complete training data set be obtained, but also a fault diagnosis model for strongly non-stationary operating conditions can be established.
  • the method architecture and training process of SSDA are shown in Figures 3 and 4 respectively.
  • the main purpose of the standard self-learning step is to train a model that can perform fault diagnosis. Since the judgment criterion of a perturbed sample is whether it can interfere with the model judgment, the judgment of the model will be regarded as the evaluation criterion and applied in the data enhancement step.
  • the model (1-D-CNN) passes the training data set come for training, among which, is the number of samples in the data set, xi represents the i-th sample in the data set, represents its label, y i is a one-hot vector, and the assignment rules for its elements are:
  • r ⁇ [0,1,2,...,R] represents the number of cycles of adversarial training, and R is the total number of cycles.
  • training data set By training data set and the data set generated by the rth data augmentation composed, and is the original training data set.
  • the model is trained through the cross-entropy objective function, which is defined as:
  • o i ⁇ c ( ⁇ f ( xi )).
  • the model uses the adaptive moment estimation algorithm (Adam) as the optimizer.
  • the number of iterations of backpropagation is recorded as T s , and the learning rate is ⁇ s ; by minimizing L s ( ⁇ ), the model will have the ability to predict the data set The ability to make a correct diagnosis on the samples in the sample.
  • Adam adaptive moment estimation algorithm
  • Generating perturbation samples is the goal of data enhancement, and the standard is whether the generated samples can interfere with the judgment of the model; therefore, using the output of the model itself as a guide, perturbation samples are generated through sample parameterization and model digitization.
  • Sample parameterization that is, treating samples as model parameters, training parameters that reduce the objective function through stochastic gradient descent, and then exporting the parameters into generated samples; model data is converted into parameters ⁇ of the 1-D-CNN model Treat as data, i.e., the parameters ⁇ are fixed during training.
  • the criterion for perturbing a sample is that it will cause perturbation to the posterior probability of the model after being input into the model. Therefore, the first objective function of data enhancement is:
  • Formula (8) shows that the objective function of data enhancement is antagonistic to the objective function of standard self-learning, so a complete data set and diagnostic model can be obtained at the same time.
  • the final objective function of the data enhancement process is:
  • parameters Adam is also used as the optimizer, the number of iterations of backpropagation is recorded as T g , and the learning rate is ⁇ g ; by minimizing parameter data set will be converted into a data set different from its initialization perturbed sample data set.
  • Step S3 Input the collected vibration signals to be diagnosed into the trained intelligent fault diagnosis model to obtain the bearing fault diagnosis results.
  • the output of the intelligent fault diagnosis model is the posterior probability that the vibration signal to be diagnosed belongs to each fault type.
  • the probabilities are sorted.
  • the fault type with the highest probability is the final bearing fault diagnosis result.
  • a motor-driven bearing failure test bench under strong non-stationary working conditions was selected for verification experiments.
  • the test bench and faulty parts are shown in Figure 5.
  • the test bench consists of a motor, tachometer, coupling, bearing seat, and double-disc rotor; the target faulty bearing is the end bearing, model NU205EM, and the acceleration sensor (PCB315A) is placed on the end bearing seat; there are three preset bearings Single faults: inner ring fault (IF), rolling element fault (RF) and outer ring fault (OF), and a compound fault: outer ring and rolling element combined fault (ORF).
  • the motor speed range is 0 ⁇ 1500rpm, and the vibration signal is collected using the LMS data acquisition system at a sampling frequency of 12.8kHz.
  • the data contains the following three forms of working conditions.
  • Constant speed working condition The speed change rate of the constant speed sample is 0. Compared with the strongly non-stationary working condition, it is more different from the training sample. It can be considered that all samples in the constant speed working condition are disturbance samples; Experiment Data at 800rpm, 1000rpm and 1500rpm speeds (represented by TD1, TD2 and TD3 respectively) were collected to test the validity of the generated data.
  • the undetermined parameters of the model T s , T g , ⁇ s , ⁇ g , ⁇ , R, and E m are preset to 100, 100, 0.01, 1, 1, 10, and 2000 respectively; after the model is trained using the incomplete training data set, it is used TD1, TD2, TD3 and TDR data sets for testing.
  • a 1-D-CNN model with the same structure as the 1-D-CNN of this method is used, and only training samples are used for training and test data are diagnosed. For comparison, the results are shown in Figure 7 Show.
  • the diagnostic accuracy of the two methods for diagnosing the TD1, TD2 and TD3 data sets is significantly less than the accuracy of diagnosing TDR. This is because the constant speed data set has more disturbance samples compared to the training data set. ; Although the model structures of 1-D-CNN and SSDA are exactly the same in the fault diagnosis process, there is a significant gap in the diagnosis results of the two methods; when 1-D-CNN diagnoses the constant speed data set, the accuracy is less than 90%.
  • the accuracy rate is only 91.67% to 92.54% when diagnosing strong non-stationary working condition data sets; compared with 1-D-CNN, the SSDA method proposed by the present invention has an accuracy rate increased by 10% when diagnosing constant rotation speed data sets Above, the accuracy of TDR has been increased to 98.55% to 99.07%, which shows that the proposed method can generate perturbation samples to expand the data set and make it closer to the complete data set.
  • This embodiment discloses a fault diagnosis system based on standard self-learning data enhancement
  • a fault diagnosis system based on standard self-learning data enhancement includes a model building module, a model training module and a fault diagnosis module;
  • the model building module is configured to: build a fault diagnosis model based on a one-dimensional convolutional neural network
  • the model training module is configured to: train the fault diagnosis model through the cross-adversarial training method of standard self-learning and data enhancement to obtain a complete data set and an intelligent fault diagnosis model under strongly non-stationary operating conditions;
  • the fault diagnosis module is configured to input the collected vibration signals to be diagnosed into the trained intelligent fault diagnosis model to obtain bearing fault diagnosis results.
  • the purpose of this embodiment is to provide a computer-readable storage medium.
  • a computer-readable storage medium has a computer program stored thereon, and when executed by a processor, the program implements the steps in a fault diagnosis method based on standard self-learning data enhancement as described in Embodiment 1 of the present disclosure.
  • the purpose of this embodiment is to provide electronic equipment.
  • An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor.
  • the processor executes the program, it implements a standard-based self-learning data enhancement as described in Embodiment 1 of the present disclosure. steps in the troubleshooting method.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Acoustics & Sound (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

一种基于标准自学习数据增强的故障诊断方法及系统，包括：基于一维卷积神经网络，构建故障诊断模型（S1）；通过标准自学习与数据增强的交叉对抗训练方式，对故障诊断模型进行训练，得到完备数据集和强非平稳工况下的智能故障诊断模型(S2)；将采集的待诊断振动信号，输入到训练好的智能故障诊断模型中，得到轴承故障诊断结果(S3)。以一维卷积神经网络为基本框架，利用不完备的训练数据集，通过标准自学习与数据增强的交叉对抗训练方式，生成扰动数据，获得强非平稳工况下的故障诊断模型，提高故障诊断的准确率。

Description

一种基于标准自学习数据增强的故障诊断方法及系统
本发明要求于2022年8月30日提交中国专利局、申请号为202211060285.9、发明名称为“一种基于标准自学习数据增强的故障诊断方法及系统”的中国专利申请的优先权，其全部内容通过引用结合在本发明中。
技术领域
本发明属于轴承故障诊断技术领域，尤其涉及一种基于标准自学习数据增强的故障诊断方法及系统。
背景技术
本部分的陈述仅仅是提供了与本发明相关的背景技术信息，不必然构成在先技术。
滚动轴承作为应用最广泛的旋转部件，是高端装备内部运动转换以及功率传输的核心；滚动轴承常常在强非平稳工况下运行，过程中载荷、转速的剧烈波动，一方面导致滚动轴承故障频发，另一方面加速损伤扩展从而加重故障危害；因此，强非平稳工况下滚动轴承故障诊断，对于保障高端装备安全、高效运行具有重要意义。
对健康监测设备捕捉的动态信号进行处理与分析，是诊断滚动轴承故障最常用的手段；随着健康监测朝着高精度、多方位、全时长方向的发展，现代健康监测设备采集了海量动态信号，进而使故障诊断进入了“大数据”时代，导致传统的基于信号分析的故障诊断方法难以满足诊断效率的要求，这催生了基于“大数据”驱动的深度学习智能故障诊断方法；同时，随着设备复杂度的提高，强非平稳工况下采集的信号中不仅伴有极强的噪声，而且故障冲击特征与其他分量还存在强烈的耦合、重叠、畸变等现象，大大增加了信号分析难度；因此，在强非平稳工况下，深度学习智能故障诊 断方法的需求更加迫切。
结构的加深意味着提取精准的故障特征更加容易，同时也造成了模型对于训练数据的过拟合，这就突显出完备训练数据的重要性；对于欠完备的健康监测数据，为提取目标特征进行模型结构的加深，只能使其陷入局限性诊断知识的过拟合，无法满足实际的诊断需求，因此，完备健康监测训练数据是智能故障诊断方法实施的基本前提。
强非平稳工况下的完备训练数据集，要求故障、瞬时工况(转速、载荷等)、工况变化率(转速、载荷变化率等)三维连续信息的叠加完备性，即每一种故障都要在任意瞬时工况、任意工况变化率下采集样本，如此苛刻的要求在实际中是无法实现的；在实际中，设备一旦发现故障，为预防严重事故必须停机检修，故障样本只是一段匀减速的动态信号，其中的工况变化率信息极为单一，而且必然缺失一定范围的瞬时工况信息(例如转速)，这远远满足不了完备性的要求；可见，强非平稳工况下采集的训练数据极不完备，这严重制约了智能故障诊断的发展。
数据增强(Data Augmentation，DA)是一种通过生成新的训练样本来处理不完备数据集最直接的方法；传统的方法起源于图像识别前处理，例如图像旋转、放大等；近年来，生成对抗神经网络(Generative Adversarial Networks，GAN)作为一种数据智能生成方法，成为数据增强的热点；在旋转机械智能故障诊断领域也提出了一些基于GAN的数据增强方法；Zhou等设计了GAN的生成器与鉴别器，采用全局优化的方案产生更多的样本来处理数据不平衡问题；Shao等和Guo等分别开发了一种基于GAN的辅助分类器框架和多标签一维GAN，从机械传感器信号中学习并生成更接近于真实的数据，来解决数据不足的问题。
现有的数据增强方法主要针对数据集不平衡、数据量小等问题，通过生成更接近于原始数据的样本来扩充数据量，进而提高模型的诊断准确率。但是，以数据相似性为目的的数据生成只能得到趋同的数据；通常情况下， 强非平稳工况运行时的旋转机械健康监测数据只是局限性的、信息缺失的匀减速数据集，一味追求生成数据的相似性只能扩充数据量，而无法弥补数据集欠缺的信息；只有生成多样性的样本，才能使强非平稳工况下的数据集满足三维连续信息的叠加完备性。因此，数据生成的重点是生成数据与原数据的差异性。
发明内容
为克服上述现有技术的不足，本发明提供了一种基于标准自学习数据增强的故障诊断方法及系统，以一维卷积神经网络为基本框架，利用不完备的训练数据集，通过标准自学习与数据增强的交叉对抗训练方式，生成扰动数据，获得强非平稳工况下的故障诊断模型，提高故障诊断的准确率。
为实现上述目的，本发明的一个或多个实施例提供了如下技术方案：
本发明第一方面提供了一种基于标准自学习数据增强的故障诊断方法；
一种基于标准自学习数据增强的故障诊断方法，包括：
基于一维卷积神经网络，构建故障诊断模型；
通过标准自学习与数据增强的交叉对抗训练方式，对故障诊断模型进行训练，得到完备数据集和强非平稳工况下的智能故障诊断模型；
将采集的待诊断振动信号，输入到训练好的智能故障诊断模型中，得到轴承故障诊断结果。
进一步的，所述一维卷积神经网络，包括多层卷积层、池化层和全连接层；
所述卷积层，使用ReLU(Rectified Linear Unit)作为激活函数，卷积操作的步幅均为1；
所有卷积层后均连接池化层，对卷积层的输出特征进行降维；
输入样本经过多层卷积与池化后的特征展平为一维向量，然后通过三层全连接进行故障诊断。
进一步的，所述标准自学习，以学习分类知识为目标，通过反复输入 更新后的样本来优化故障诊断模型中的参数，自学判断样本是否为扰动样本的评价标准。
进一步的，所述数据增强，以模型本身的输出为指导，通过样本参数化与模型数据化的方法生成扰动样本；
其中，扰动样本的判断标准是其能否干扰模型判断，具体为：样本输入到模型后能引起模型后验概率的扰动。
进一步的，所述样本参数化，是将样本看作模型参数，通过随机梯度下降法训练出使目标函数降低的参数，进而将参数导出为生成的样本。
进一步的，所述模型数据化，是将故障诊断模型的参数看作数据，在训练过程中固定参数值。
进一步的，所述智能故障诊断模型的输出，是待诊断振动信号属于各个故障类型的后验概率，对概率进行排序，概率最高的故障类型，是最终的轴承故障诊断结果。
本发明第二方面提供了一种基于标准自学习数据增强的故障诊断系统。
一种基于标准自学习数据增强的故障诊断系统，包括模型构建模块、模型训练模块和故障诊断模块；
模型构建模块，被配置为：基于一维卷积神经网络，构建故障诊断模型；
模型训练模块，被配置为：通过标准自学习与数据增强的交叉对抗训练方式，对故障诊断模型进行训练，得到完备数据集和强非平稳工况下的智能故障诊断模型；
故障诊断模块，被配置为：将采集的待诊断振动信号，输入到训练好的智能故障诊断模型中，得到轴承故障诊断结果。
本发明第三方面提供了计算机可读存储介质，其上存储有程序，该程序被处理器执行时实现如本发明第一方面所述的一种基于标准自学习数据增强的故障诊断方法中的步骤。
本发明第四方面提供了电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的程序，所述处理器执行所述程序时实现如本发明第一方面所述的一种基于标准自学习数据增强的故障诊断方法中的步骤。
以上一个或多个技术方案存在以下有益效果：
本发明提出了一种标准自学习数据增强方法，以故障诊断模型自身预测结果为数据生成的评价标准，通过样本参数化与模型数据化来生成扰动样本，扩充数据集使其更接近完备数据集。
本发明以一维卷积神经网络为基本框架，利用不完备的训练数据集，通过标准自学习与数据增强的交叉对抗训练方式，获得强非平稳工况下的故障诊断模型，提高故障诊断的准确率。
本发明附加方面的优点将在下面的描述中部分给出，部分将从下面的描述中变得明显，或通过本发明的实践了解到。
附图说明
构成本发明的一部分的说明书附图用来提供对本发明的进一步理解，本发明的示意性实施例及其说明用于解释本发明，并不构成对本发明的不当限定。
图1(a)-(b)为被不完备数据集训练的人类识别常规目标和识别扰动样本的示例图。
图2为第一个实施例的方法流程图。
图3为标准自学习数据增强方法的架构图。
图4为标准自学习数据增强方法的训练流程图。
图5为故障实验台及故障轴承。
图6为TDR数据集中不同健康状况样本的转速变化情况。
图7为不同测试集的诊断结果。
图8为第二个实施例的系统结构图。
具体实施方式
下面结合附图与实施例对本发明作进一步说明。
应该指出，以下详细说明都是示例性的，旨在对本发明提供进一步的说明。除非另有指明，本文使用的所有技术和科学术语具有与本发明所属技术领域的普通技术人员通常理解的相同含义。
需要注意的是，这里所使用的术语仅是为了描述具体实施方式，而非意图限制根据本发明的示例性实施方式。如在这里所使用的，除非上下文另外明确指出，否则单数形式也意图包括复数形式，此外，还应当理解的是，当在本说明书中使用术语“包含”和/或“包括”时，其指明存在特征、步骤、操作、器件、组件和/或它们的组合。
实施例一
人类的目标识别常常被不完备的训练数据集所困扰，如图1所示，一个仅见过常规鱼类的人，在看到“鲤鱼”、“草鱼”等鱼类时，能立刻识别出来；但当其看到“飞鱼”时，脑中或许在犹豫目标是鱼还是鸟，而且极有可能识别错误。
上述例子中，仅见过常规鱼类的人相当于被不完备数据集训练的模型；“鲤鱼”、“草鱼”等鱼类可以看作是与训练数据相似的样本，本实施例中称为常规样本；“飞鱼”相当于与训练样本不相似的样本(本实施例中称为扰动样本)；可见，被不完备数据集训练的模型，识别扰动样本的能力大大降低。
强非平稳工况下采集的故障数据集是典型的不完备数据集，但模型的任务往往是在复杂多变的工况下进行诊断，这就意味着测试样本大多是扰动样本；因此，数据增强的目的是生成扰动样本来扩充训练集，进而增强训练集的完备性；基于GAN的方法，其目的是生成与训练样本近似的常规样本，这显然与强非平稳工况下数据增强的目的相悖；因此，亟需转变常规智能数据增强思路，进而提出面向扰动数据的智能数据增强方法。
若要生成扰动样本，首要问题是阐明扰动样本的标准，即，如何评价生成的样本是扰动样本。从上述例子中可以看出，“飞鱼”作为扰动样本可以使人脑的预测结果发生错误，这就说明，只有当样本与原始训练集的差异化达到足以干扰模型判断时，才可以视其为扰动样本，受此思想启发，本发明提出了以故障诊断模型自身预测结果为数据生成的评价标准，通过样本参数化与模型数据化来生成样本的数据增强方法；由于故障诊断模型自身预测结果是模型通过训练数据学习出来的，因此称为标准自学习数据增强方法SSDA(Standard Self-learned Data Augmentation)。
本实施例公开了一种基于标准自学习数据增强的故障诊断方法，如图2所示，具体包括：
步骤S1、基于一维卷积神经网络，构建故障诊断模型；
SSDA的基本模型架构采用了目前广泛使用的一维卷积神经网络(One-Dimensional Convolutional Neural Networks，1-D-CNN)；1-D-CNN包括多层卷积层、池化层和全连接层，各层参数集如表1所示：
表1 1-D-CNN的逐层参数
为了减少人工工作量，原始测量的振动信号经过分段后，无需经过傅 里叶变换等信号处理方法，直接输入到网络中；所构建模型的输入样本定义为其中N为样本维度，本研究将N设为1200维。
卷积层
对于第l层卷积层，其特征可通过下式获得：
其中，为卷积核，且Kl为卷积核的长度，Ml-1为前一特征层的通道数，Ml为当前层的通道数；是前一特征层的输出特征，是Nl-1×Ml-1的向量空间，且Nl-1为特征维数；bl为偏置向量；f(·)为激活函数，本实施例中卷积层全部使用ReLU(Rectified Linear Unit)作为激活函数；vl-1*kl为卷积操作，可通过下式计算：
其中，下标[·]代表矩阵中元素的序号，卷积操作的步幅均为1，因此，的维度为(Nl-1-Kl+1)×Ml。
池化层
本模型的所有卷积层后均连接池化层，对第l层卷积层的输出特征进行降维，池化层的输出特征vl为：
其中，S为池化长度。
全连接层
输入样本经过多层卷积与池化后的特征展平为一维向量u1，然后通过三层全连接进行故障诊断，全连接的前向传播为：
ul＝f(wlul-1+bl)……(4)
其中，wl和bl分别为全连接层的权值矩阵和偏置向量；前两层全连接层的激活函数为ReLU激活函数，最后一层全连接层特征经过Softmax激活函数得到模型的输出(C代表故障类型数)，即，输出o中的元素可通过下式计算：
其中，代表w3u3+b3，即输出层未经过激活函数的特征。
模型的输出o代表样本属于各个故障类型的后验概率，因此根据模型输出即可判断样本的故障类型，为了方便描述，样本x输入模型后转化为特征u1的过程抽象为映射Φf，特征u1转化为输出o的过程抽象为映射Φc，即u1＝Φf(x)，o＝Φc(u1)；模型的所有参数用θ表示。
步骤S2、通过标准自学习与数据增强的交叉对抗训练方式，对故障诊断模型进行训练，得到完备数据集和强非平稳工况下的智能故障诊断模型；
SSDA包含标准自学习和数据增强两个训练步骤；在标准自学习中，以学习分类知识为目标，通过反复输入更新后的样本来优化1-D-CNN模型中的参数，此过程相当于模型自学了判断样本是否为扰动样本的评价标准；在数据增强中，通过样本参数化与模型数据化的方法，对模型输出结果的后验概率进行干扰，从而产生多样化的样本；经过两个训练步骤的交替进行，最终不仅能获得完备的训练数据集，还能建立面向强非平稳工况的故障诊断诊断模型，SSDA的方法架构和训练流程分别如图3、4所示。
标准自学习
标准自学习步骤的主要目的是训练出能够进行故障诊断的模型，由于扰动样本的判断标准是其能否干扰模型判断，因此，模型的判断将被视为评价标准应用于数据增强步骤中。
模型(1-D-CNN)通过训练数据集来训练的，其中，为数据集中的样本个数，xi表示数据集中的第i个样本，代表其标签，yi为独热向量，其元素的赋值规则为：
参数中，r∈[0,1,2,…,R]代表对抗训练的循环次数，R为总循环次数。 训练数据集由训练数据集和第r次数据增强生成的数据集组成，且为原始的训练数据集。
在标准自学习过程中，模型通过交叉熵目标函数进行训练，其定义为：
其中，oi＝Φc(Φf(xi))。
模型采用自适应矩估计算法(Adam)作为优化器，反向传播的迭代次数记为Ts，学习率为εs；通过最小化Ls(θ)，模型将具备对数据集中的样本进行正确诊断的能力。
数据增强
生成扰动样本是数据增强的目标，其标准是生成的样本能否干扰模型的判断；因此，以模型本身的输出为指导，通过样本参数化与模型数据化的方法生成扰动样本。
样本参数化，即，将样本看作模型参数，通过随机梯度下降法训练出使目标函数降低的参数，进而将参数导出为生成的样本；模型数据化为将1-D-CNN模型的参数θ看作数据，即，在训练过程中固定参数θ。
因此，首先以数据集为初值初始化参数其中，为上一次生成的样本，且这意味着模型在之前生成的样本的基础上进一步实施数据增强。
扰动样本的标准是其输入模型后将引起模型后验概率的扰动，因此数据增强的第一项目标函数为：
其中，且 代表原初始化数据集中样本的个数；公式(8)说明数据增强的目标函数与标准自学习的目标函数是对抗的，因此能同时得到完备的数据集和诊断模型。
若数据增强过程只关注扰动性，则容易使后验概率偏差过大而生成无 意义的样本，对样本生成过程加以限制是有必要的。因此，数据增强的第二项目标函数为
其中，u1,i＝Φf(xi)，λ＞0为调节系数；参与优化意味着数据增强过程限制了过大的样本变化，但允许合理的样本多样性存在。
数据增强过程最终的目标函数为：
其中，参数同样采用Adam作为优化器，反向传播的迭代次数记为Tg，学习率为εg；通过最小化参数数据集将转化为不同于其初始化数据集的扰动样本数据集。
训练策略
如图4所示，在标准自学习数据增强故障诊断方法中，标准自学习与数据增强过程交替进行，从而得到完备数据集和强非平稳工况下的智能故障诊断模型，具体训练流程如下：
(1)初始化数据集并随机初始化模型参数θ0，设置超参数Ts、Tg、εs、εg、λ、R以及对抗循环R次以后模型的额外训练次数Em，初始化r＝0。
(2)基于训练数据集进行标准自学习，直至达到最大迭代次数Ts，令r＝r+1，进而得到训练的模型参数θr。
(3)采用模型数据化与样本参数化方法进行数据增强，利用数据集经过Tg次迭代，生成新的扰动样本数据集
(4)合并与组成新的训练集
(5)判断是否达到最大循环次数R，如果r＜R，返回第(2)步。否则，基于训练集进行标准自学习直至达到额外训练次数Em。
(6)完成训练，获得完备数据集和具有最优参数集θR+1的用于强非平稳工况下的故障诊断模型。
步骤S3、将采集的待诊断振动信号，输入到训练好的智能故障诊断模型中，得到轴承故障诊断结果。
智能故障诊断模型的输出，是待诊断振动信号属于各个故障类型的后验概率，对概率进行排序，概率最高的故障类型，是最终的轴承故障诊断结果。
通过实验及结果的分析，验证本发明所提出的一种基于标准自学习数据增强的故障诊断方法在强非平稳工况下的准确率。
数据描述
选用电机驱动的强非平稳工况轴承故障实验台进行验证实验，实验台与故障零件如图5所示。试验台由电动机、转速计、联轴器、轴承座、双盘转子组成；目标故障轴承为端部轴承，型号为NU205EM，加速度传感器(PCB315A)放置于端部轴承座上；轴承预设三种单一故障：内圈故障(I F)、滚动体故障(RF)和外圈故障(OF)，以及一种复合故障：外圈与滚动体复合故障(ORF)。电机转速范围为0～1500rpm，振动信号采用LMS数据采集系统，以12.8kHz采样频率采集。
为验证所提方法的有效性，数据共包含以下三种形式的工况。
(1)匀减速工况：电机由1500rpm匀减速至静止，此过程模拟实际运转中出现故障停机时采集到的不完备数据集，为本方法的训练数据。
(2)强非平稳工况：此工况模拟实际运行中设备的强非平稳工况，转速变化情况如图6所示，作为验证本方法的测试数据，用TDR表示。
(3)恒转速工况：恒转速样本的转速变化率为0，相对于强非平稳工况，其与训练样本的差异性更大，可以认为恒转速工况的样本全部为扰动样本；实验采集了800rpm、1000rpm和1500rpm转速下的数据(分别用TD1、TD2和TD3表示)来测试生成数据的有效性。
实验结果分析
模型待定参数Ts、Tg、εs、εg、λ、R以及Em分别预设置为100、100、0.01、1、1、10、2000；模型利用不完备训练数据集训练后，采用TD1、TD2、TD3和TDR数据集进行测试。为了验证所提方法的有效性，采用与本方法1-D-CNN相同结构的1-D-CNN模型，仅使用训练样本进行训练并对测试数据进行诊断，作为对比，所得结果如图7所示。
图7中可以看出，两种方法诊断TD1、TD2和TD3数据集的诊断准确率明显小于诊断TDR的准确率，这是由于相比于训练数据集，恒转速数据集具有更多的扰动样本；尽管1-D-CNN与SSDA在故障诊断过程中的模型结构完全相同，但两种方法的诊断结果有显著差距；1-D-CNN在诊断恒转速数据集时，准确率均不足90％，诊断强非平稳工况数据集时也仅有91.67％～92.54％的准确率；本发明所提出的SSDA方法相比1-D-CNN，诊断恒转速数据集时，准确率提高了10％以上，对TDR的准确率提高到了98.55％～99.07％，这说明提出的方法能够生成扰动样本来扩充数据集使其更接近完备数据集。
实施例二
本实施例公开了一种基于标准自学习数据增强的故障诊断系统；
如图8所示，一种基于标准自学习数据增强的故障诊断系统，包括模型构建模块、模型训练模块和故障诊断模块；
模型构建模块，被配置为：基于一维卷积神经网络，构建故障诊断模型；
模型训练模块，被配置为：通过标准自学习与数据增强的交叉对抗训练方式，对故障诊断模型进行训练，得到完备数据集和强非平稳工况下的智能故障诊断模型；
故障诊断模块，被配置为：将采集的待诊断振动信号，输入到训练好的智能故障诊断模型中，得到轴承故障诊断结果。
实施例三
本实施例的目的是提供计算机可读存储介质。
计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如本公开实施例1所述的一种基于标准自学习数据增强的故障诊断方法中的步骤。
实施例四
本实施例的目的是提供电子设备。
电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的程序，所述处理器执行所述程序时实现如本公开实施例1所述的一种基于标准自学习数据增强的故障诊断方法中的步骤。
以上所述仅为本发明的优选实施例而已，并不用于限制本发明，对于本领域的技术人员来说，本发明可以有各种更改和变化。凡在本发明的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种基于标准自学习数据增强的故障诊断方法，其特征在于，包括：
    基于一维卷积神经网络，构建故障诊断模型；
    通过标准自学习与数据增强的交叉对抗训练方式，对故障诊断模型进行训练，得到完备数据集和强非平稳工况下的智能故障诊断模型；
    将采集的待诊断振动信号，输入到训练好的智能故障诊断模型中，得到轴承故障诊断结果。
  2. 如权利要求1所述的一种基于标准自学习数据增强的故障诊断方法，其特征在于，所述一维卷积神经网络，包括多层卷积层、池化层和全连接层；
    所述卷积层，使用ReLU(Rectified Linear Unit)作为激活函数，卷积操作的步幅均为1；
    所有卷积层后均连接池化层，对卷积层的输出特征进行降维；
    输入样本经过多层卷积与池化后的特征展平为一维向量，然后通过三层全连接进行故障诊断。
  3. 如权利要求1所述的一种基于标准自学习数据增强的故障诊断方法，其特征在于，所述标准自学习，以学习分类知识为目标，通过反复输入更新后的样本来优化故障诊断模型中的参数，自学判断样本是否为扰动样本的评价标准。
  4. 如权利要求1所述的一种基于标准自学习数据增强的故障诊断方法，其特征在于，所述数据增强，以模型本身的输出为指导，通过样本参数化与模型数据化的方法生成扰动样本；
    其中，扰动样本的判断标准是其能否干扰模型判断，具体为：样本输 入到模型后能引起模型后验概率的扰动。
  5. 如权利要求4所述的一种基于标准自学习数据增强的故障诊断方法，其特征在于，所述样本参数化，是将样本看作模型参数，通过随机梯度下降法训练出使目标函数降低的参数，进而将参数导出为生成的样本。
  6. 如权利要求4所述的一种基于标准自学习数据增强的故障诊断方法，其特征在于，所述模型数据化，是将故障诊断模型的参数看作数据，在训练过程中固定参数值。
  7. 如权利要求1所述的一种基于标准自学习数据增强的故障诊断方法，其特征在于，所述智能故障诊断模型的输出，是待诊断振动信号属于各个故障类型的后验概率，对概率进行排序，概率最高的故障类型，是最终的轴承故障诊断结果。
  8. 一种基于标准自学习数据增强的故障诊断系统，其特征在于，包括模型构建模块、模型训练模块和故障诊断模块；
    模型构建模块，被配置为：基于一维卷积神经网络，构建故障诊断模型；
    模型训练模块，被配置为：通过标准自学习与数据增强的交叉对抗训练方式，对故障诊断模型进行训练，得到完备数据集和强非平稳工况下的智能故障诊断模型；
    故障诊断模块，被配置为：将采集的待诊断振动信号，输入到训练好的智能故障诊断模型中，得到轴承故障诊断结果。
  9. 计算机可读存储介质，其上存储有程序，其特征在于，该程序被处理器执行时实现如权利要求1-7任一项所述的一种基于标准自学习数据增 强的故障诊断方法中的步骤。
  10. 电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的程序，其特征在于，所述处理器执行所述程序时实现如权利要求1-7任一项所述的一种基于标准自学习数据增强的故障诊断方法中的步骤。
PCT/CN2023/081715 2022-08-30 2023-03-15 一种基于标准自学习数据增强的故障诊断方法及系统 WO2024045555A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211060285.9 2022-08-30
CN202211060285.9A CN115753103A (zh) 2022-08-30 2022-08-30 一种基于标准自学习数据增强的故障诊断方法及系统

Publications (1)

Publication Number Publication Date
WO2024045555A1 true WO2024045555A1 (zh) 2024-03-07

Family

ID=85349490

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/081715 WO2024045555A1 (zh) 2022-08-30 2023-03-15 一种基于标准自学习数据增强的故障诊断方法及系统

Country Status (2)

Country Link
CN (1) CN115753103A (zh)
WO (1) WO2024045555A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115753103A (zh) * 2022-08-30 2023-03-07 山东建筑大学 一种基于标准自学习数据增强的故障诊断方法及系统

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110608884A (zh) * 2019-08-08 2019-12-24 桂林电子科技大学 一种基于自注意力神经网络的滚动轴承状态诊断方法
CN111539152A (zh) * 2020-01-20 2020-08-14 内蒙古工业大学 一种基于两级孪生卷积神经网络的滚动轴承故障自学习方法
US20200285900A1 (en) * 2019-03-06 2020-09-10 Wuhan University Power electronic circuit fault diagnosis method based on optimizing deep belief network
CN112067294A (zh) * 2019-09-20 2020-12-11 宫文峰 基于深度学习的滚动轴承智能故障诊断方法
CN112284735A (zh) * 2020-10-21 2021-01-29 兰州理工大学 基于一维卷积与动态路由的多传感器滚动轴承故障诊断
CN113095413A (zh) * 2021-04-14 2021-07-09 山东建筑大学 变工况故障诊断方法、系统、存储介质及设备
CN113203566A (zh) * 2021-04-06 2021-08-03 上海吞山智能科技有限公司 一种基于一维数据增强和cnn的电机轴承故障诊断方法
US20210278478A1 (en) * 2020-03-02 2021-09-09 Wuhan University Deep parallel fault diagnosis method and system for dissolved gas in transformer oil
CN114295377A (zh) * 2021-12-13 2022-04-08 南京工业大学 一种基于遗传算法的cnn-lstm轴承故障诊断方法
WO2022077901A1 (zh) * 2020-10-13 2022-04-21 同济大学 一种面向小样本数据集的轴承故障模式诊断方法及系统
US20220269925A1 (en) * 2019-06-05 2022-08-25 South China University Of Technology Intelligent fault diagnosis method based on multi-task feature sharing neural network
CN115753103A (zh) * 2022-08-30 2023-03-07 山东建筑大学 一种基于标准自学习数据增强的故障诊断方法及系统

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200285900A1 (en) * 2019-03-06 2020-09-10 Wuhan University Power electronic circuit fault diagnosis method based on optimizing deep belief network
US20220269925A1 (en) * 2019-06-05 2022-08-25 South China University Of Technology Intelligent fault diagnosis method based on multi-task feature sharing neural network
CN110608884A (zh) * 2019-08-08 2019-12-24 桂林电子科技大学 一种基于自注意力神经网络的滚动轴承状态诊断方法
CN112067294A (zh) * 2019-09-20 2020-12-11 宫文峰 基于深度学习的滚动轴承智能故障诊断方法
CN111539152A (zh) * 2020-01-20 2020-08-14 内蒙古工业大学 一种基于两级孪生卷积神经网络的滚动轴承故障自学习方法
US20210278478A1 (en) * 2020-03-02 2021-09-09 Wuhan University Deep parallel fault diagnosis method and system for dissolved gas in transformer oil
WO2022077901A1 (zh) * 2020-10-13 2022-04-21 同济大学 一种面向小样本数据集的轴承故障模式诊断方法及系统
CN112284735A (zh) * 2020-10-21 2021-01-29 兰州理工大学 基于一维卷积与动态路由的多传感器滚动轴承故障诊断
CN113203566A (zh) * 2021-04-06 2021-08-03 上海吞山智能科技有限公司 一种基于一维数据增强和cnn的电机轴承故障诊断方法
CN113095413A (zh) * 2021-04-14 2021-07-09 山东建筑大学 变工况故障诊断方法、系统、存储介质及设备
CN114295377A (zh) * 2021-12-13 2022-04-08 南京工业大学 一种基于遗传算法的cnn-lstm轴承故障诊断方法
CN115753103A (zh) * 2022-08-30 2023-03-07 山东建筑大学 一种基于标准自学习数据增强的故障诊断方法及系统

Also Published As

Publication number Publication date
CN115753103A (zh) 2023-03-07

Similar Documents

Publication Publication Date Title
Jiang et al. A GAN-based anomaly detection approach for imbalanced industrial time series
CN111914883B (zh) 一种基于深度融合网络的主轴轴承状态评估方法及装置
CN111046945B (zh) 基于组合卷积神经网络的故障类型及损坏程度诊断方法
CN109489946A (zh) 一种旋转机械的故障诊断方法及系统
CN112257530B (zh) 基于盲信号分离和支持向量机的滚动轴承故障诊断方法
CN113485302B (zh) 基于多元时序数据的车辆运行过程故障诊断方法及系统
Yang et al. Fault diagnosis of reciprocating compressor based on convolutional neural networks with multisource raw vibration signals
WO2024045555A1 (zh) 一种基于标准自学习数据增强的故障诊断方法及系统
Di et al. Ensemble deep transfer learning driven by multisensor signals for the fault diagnosis of bevel-gear cross-operation conditions
CN112729834A (zh) 一种轴承故障的诊断方法、装置和系统
CN115587290A (zh) 基于变分自编码生成对抗网络的航空发动机故障诊断方法
CN115409052A (zh) 一种数据失衡下风力发电机组轴承的故障诊断方法及系统
CN114881071A (zh) 基于多源信息的同步电机转子绕组匝间短路故障诊断方法
Liu et al. Rolling bearing fault diagnosis of train running gear based on optimized deep residual network
CN111783941B (zh) 一种基于概率置信度卷积神经网络的机械设备诊断分类方法
CN117349595A (zh) 一种面向工业场景的极端少样本故障诊断方法
CN116625686A (zh) 一种航空发动机轴承故障在线诊断方法
CN115795397A (zh) 基于1DCNN-LSTM与BiLSTM并行网络的齿轮箱温度预测方法
CN115556099B (zh) 一种可持续学习的工业机器人故障诊断系统及方法
CN114877925B (zh) 一种基于极限学习机的综合能源系统传感器故障诊断方法
Yu et al. A new multi-source information domain adaption network based on domain attributes and features transfer for cross-domain fault diagnosis
Pan et al. Generalized Zero-shot Learning Fault Diagnosis Framework Based on Anomaly Detection and Contractive Stacked Autoencoder
CN114383846B (zh) 基于故障标签信息向量的轴承复合故障诊断方法
CN114580284B (zh) 旋转机械变工况故障诊断方法及系统
Li et al. Intelligent CNN-based Fault Diagnosis of Rotating Machinery with Small Fault Samples

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: 23858597

Country of ref document: EP

Kind code of ref document: A1