Search

Penetration recognition is the critical technology to improve manufacturing quality and automation level in arc welding. In this paper, recent research advances concerning penetration recognition based on machine learning are comprehensively reviewed, focused on signals feature extraction and application of machine learning in this field. Three types of signals originating from arc welding, namely, weld pool image, arc sound, and arc voltage, are used to analyze the correlation with the penetration state and extract effective features. Next, the following contents briefly introduce some machine learning methods including conventional machine learning and deep learning, and emphatically summarize their application and performance in penetration recognition. Notably, the deep learning method possesses higher classification accuracy and better generalization ability in penetration recognition due to its deeper architecture and effective learning capacity. In the end, some challenges and tendencies, involving multi-sensor information fusion, data-driven recognition, model development incorporating attention mechanism, lightweight model deployment, and real-time control, are presented for further research. This paper is an attempt to provide a reference source and some guidance for researchers who use machine learning methods for penetration monitoring. [ABSTRACT FROM AUTHOR]

In milling processing, the wear state of the tool has an essential influence on the processing quality. The machining process is not continuous in the cycloid milling process, and the signal of the empty tool part increases the difficulty of identifying the tool wear state. At present, most of the researchers use experimental data and cut out the signal of the empty tool part in the signal by data processing. However, it will affect the original signal to a certain extent and destroy the confidential information in the original signal. A novel method using variational auto-encoder (VAE) for tool wear status identification is proposed. Due to VAE has structural characteristics that reduce the dimensionality of high-dimensional data to lower dimensionality. This requires VAE to find and learn the significant features, which are hidden in the complex raw data. In this paper, the signals of the empty tool do not need to be cut out; the effective value of the three-phase current signals obtained in the real processing is converted into the form of three-dimensional color images. VAE is applied to extract features from the image samples and then realize the classification of different wear states of the tool. A large number of comparison experiments are conducted, and the result shows that the presented method has a better recognition performance for the actual processing data. It is more suitable for the recognition of tool wear status in the actual milling process. [ABSTRACT FROM AUTHOR]

Among the diverse challenges in machining processes, chatter has a significant detrimental effect on surface quality and tool life, and it is a major limitation factor in achieving higher material removal rate. Early detection of chatter occurrence is considered a key element in the milling process automation. Online detection of chatter onset has been continually investigated over several decades, along with the development of new signal processing and machining condition classification approaches. This paper presents a review of the literature on chatter detection in milling, providing a comprehensive analysis of the reported methods for sensing and testing parameter design, signal processing and various features proposed as chatter indicators. It discusses data-driven approaches, including the use of different techniques in the time–frequency domain, feature extraction, and machining condition classification. The review outlines the potential of using multiple sensors and information fusion with machine learning. To conclude, research trends, challenges and future perspectives are presented, with the recommendation to study the tool wear effects, and chatter detection at dissimilar milling conditions, while utilization of considerable large datasets—Big Data—under the Industry 4.0 framework and the development of machining Digital Twin capable of real-time chatter detection are considered as key enabling technologies for intelligent manufacturing. [ABSTRACT FROM AUTHOR]

A reliable material removal rate (MRR) prediction method significantly optimizes the grinding surface quality and improves the processing efficiency for robotic abrasive belt grinding. Using worn-belt image features to predict MRR is a direct and reliable method; however, this method is rarely reported at present. This paper proposes an MRR prediction method for Inconel 718 grinding based on the abrasive belt image analysis and categorical boosting (CatBoost) algorithm. During belt grinding, four wear types of abrasive belts, namely fracture, adhesion, rubbing wear, and fall-off, are identified and analyzed. Under various grinding parameters, the experimental MRR rapidly decreases at first, then in a gradual manner. For an effective evaluation of belt wear severity, cutting grain area ratio, color moments, and texture features are extracted from belt images. MRR and abrasive belt image features are strongly correlated after normalization. All image features are taken into account for MRR prediction model training. Verification experiments indicate that the predicted data is in good agreement with the experimental data. The maximum absolute error, mean absolute error, root mean square error, and determination coefficient of the MRR prediction model are 0.17 μm, 0.4 μm, 0.2 μm, and 99.42%, respectively, which are superior to those of other popular machine learning algorithms. In this study, we present a comprehensive understanding of the relationship between MRR and abrasive belt characteristics, as well as demonstrate the feasibility of accurately predicting MRR using the CatBoost algorithm. [ABSTRACT FROM AUTHOR]

Recently, with the advance in information technology, pure data-driven approaches such as machine learnings have been widely applied in status diagnosis. However, the accuracy of those predictions strongly relies on the original data, which largely depends on the selected sensors and signal features. Furthermore, for unsupervised machine learning schemes, although it could avoid the concern of labeling in training, it lacks a quantified evaluation of the prediction results. These concerns significantly limit the effectiveness of modern machine learning and thus should be investigated. Meanwhile, ball bearings are fundamental key machine elements in rotating machinery and their condition monitoring should be critical for both quality control and longevity assessment. In this paper, by utilizing ball bearing failure diagnosis as the main theme, the flow of feature selection and evaluation, as well as the evaluation flow for multiple failure diagnosis, is developed for accessing the status of bearings in their imbalance, lubrication, and grease contamination levels based on unsupervised machine learning. The experimental results indicated that with proper feature selection, the failure identification could be more definite. Finally, a novel model based on the second norm to quantify the classification level of each cluster in hyperspace is proposed as the measure for unsupervised machine learning as the basis for performance evaluation and optimization of unsupervised machine learning schemes and should benefit related machine reliability evaluation studies and applications. [ABSTRACT FROM AUTHOR]

Chatter is a kind of self-excited vibration that often occurs in the milling process of thin-walled parts, which has become the main factor restricting production efficiency and quality. Due to the occurrence of chatter, the signal becomes more complex and unstable. In order to realize milling chatter detection of thin-walled parts, the method of multi-sensor signal fusion is used. A chatter detection method based on variational mode decomposition (VMD) and nonlinear dimensionless index is proposed by analyzing the characteristics of signals in time–frequency domain. Firstly, a series of intrinsic mode function (IMF) components are obtained by decomposing force and acceleration signals with VMD. When chatter occurs, the energy is transferred to the chatter frequency band. Each IMF signal's nonlinear energy entropy (EE) is extracted to construct the feature vector. A support vector machine chatter identification model based on multi-sensor signal fusion is established. To solve the problem of model incremental updating, supervised learning and unsupervised learning are combined to provide a method for chatter detection. [ABSTRACT FROM AUTHOR]

Machinery condition monitoring is a crucial application to enable condition-based predictive maintenance, realizing smart manufacturing in the context of the Industry 4.0 revolution. Machine learning that is built upon data correlation has recently become mainstream for machinery fault diagnosis. To boost the fault diagnosis performance of the machine learning methods, extracting high-quality fault-related features from raw time-series response signals is the key. In this research, we propose a signal processing-based feature extraction method, the so-called recurrence plot (RP), to evaluate the hidden dynamic characteristics of the vibration signals collected from the machinery system that is subject to fault. To further examine the properties of RP, the recurrence quantification analysis (RQA) is employed to quantitatively measure the patterns in the RP. The main strength of the proposed methodology lies in its excellent capability in handling the nonstationary signals with strong system nonlinearity and ambient noise, thereby yielding high-quality features with a high degree of separation in terms of the fault condition. A novel quantitative metric based on the silhouette is established to thoroughly assess the feature quality, providing systematic guidance for machinery condition monitoring. Comprehensive case studies using publicly accessible gear and bearing fault datasets are carried out to validate the proposed methodology. The enhanced performance of the proposed methodology also is highlighted by comparing it with other benchmark methods. [ABSTRACT FROM AUTHOR]

A great percentage of breakdowns in the rotary machines are caused by faulty gears and bearings generating high costly downtime. The diagnosis of gears and bearings combined faults using conventional frequency methods is difficult because of the presence of a high level of noise in the signal or a disturbance emanating from other sources in/out of the machine. This manuscript develops an improved approach for gears and bearings simple/combined faults automated diagnosis using an optimized wavelet packet transform (OWPT), for signal denoising and fault features extraction, combined with pattern recognition neural networks (PRNN) for fault classification. We applied this improved approach on acceleration signals acquired from a test rig simulating an industrial rotary machine; then, these acquired signals were processed using OWPT, based on the best choice of the wavelet packet and the level of decomposition using minimum Shannon entropy and maximum energy to Shannon entropy ratio criteria. The results gave us the bi-orthogonal 3.1 (bior3.1) as the best mother wavelet among 41 wavelet families being tested and the 6th level as the best decomposition level. These two found parameters are used for the extraction of time-domain features, which are used as input data to train neural networks after data normalization. The results showed that the proposed method could detect any type (simple/combined) and size (small "incipient," mean and large) of gears and bearig faults in non-stationary operations with the presence of high noise. In addition, the automation of the classification process could be generalized to any kind of faults in the machine such as misalignment, electrical faults, and rotor damage. [ABSTRACT FROM AUTHOR]

Machine learning has been widely used in diagnosing system faults as an integration part in modern intelligent manufacturing. However, sensor selection and index for machine tools are versatile and not standardized at this moment for manufacturing equipment. Without supporting of appropriate domain knowledge for selecting appropriate sensors and adequate performance index, pure data-driven approach might suffer from unsatisfied prediction accuracy and needing of excessive training data, as well as the possibility of misjudgment. As a result, a thoughtful flow for identifying key sensor index and to evaluate its impact on the performance of subsequent machine learning scheme would be essential. In this work, the status monitoring and prediction of a milling cutter wear problem is investigated as an example to address the above concerns and to demonstrate the possible solutions by hiring a 5-axis machine center equipped with milling cutters of different wear levels. In parallel, the well-used multilayer perception (MLP) artificial neural network (ANN) models are also used to elucidate the importance of the problem. Transducers including accelerometers, microphones, current transformer, and acoustic emission sensors are mounted on the spindle, fixture, and nearby structures to monitor the milling process. The collected data are processed to extract various signatures. A procedure is developed to evaluate the sensitivity of each index and the key dominated indexes are eventually identified. To elucidate the importance of proper sensor index selection, three MLP ANN models are established based on different sensor features to examine the influence of selected sensors indexes on the prediction accuracy. The results show that with appropriate sensors signatures, even with less amount of experimental data, the model can indeed achieve a better prediction (98.7%) in comparison with that without proper sensor index selection (90.3). This implies that a rigorous process for identifying key feature is essential for system diagnosis in future intelligent manufacturing. [ABSTRACT FROM AUTHOR]

Deep learning methods recently have gained growing interests and are extensively applied in the data-driven bearing fault diagnosis. However, current deep learning methods perform the bearing fault diagnosis in the form of deterministic classification, which overlook the uncertainties that inevitably exist in actual practice. To tackle this issue, in this research, we develop a probabilistic fault diagnosis framework that can account for the uncertainty effect in prediction, which bears practical significance. This framework uses the Gaussian process classifier (GPC) as the mainstay, which fundamentally is built upon the Bayesian inference. To establish the high-fidelity GPC, the tailored feature extraction method can be adaptively determined through the cross validation-based grid search upon a prespecified method pool consisting of various kernel principal component analysis (KPCA) methods and stacked autoencoder. This adaptive strategy can ensure the adequate GPC model training to accurately characterize the complex nonlinear relations between the data features and respective faults. Systematic case studies using the publicly accessible experimental rolling bearing dataset, i.e., CWRU bearing dataset are carried out to validate this new framework. The results clearly illustrate the unique capability of this framework in handling uncertainties. It is also found that this framework outperforms other well-established machine learning and deep learning models in terms of accuracy and robustness. Moreover, the sensor fusion that combines the spatial vibration measurements appears to be an effective technique to further enhance the fault diagnosis performance. By fully leveraging the probabilistic feature of the framework, the future research endeavor, such as the extended fault diagnosis using limited fault labels will be facilitated. [ABSTRACT FROM AUTHOR]