Search

With the advancement of intelligent manufacturing, different kinds of industrial robots have been applied in modern factories. The liquid crystal display transfer robot (LCDTR) has been widely used in LCD production lines to transport panels. Effective fault diagnosis and prognosis of the industrial robots are of great importance, since unplanned downtime caused by faulty robots significantly reduces the production capacity. Specifically, the ball screw is the critical component in the LCDTR. The failure of the ball screw can cause long downtime. Conventionally, the fault diagnosis of the ball screw is usually based on the vibration signals. However, it is extremely difficult to install the vibration sensors in the industrial robots. Therefore, in order to address this issue in condition monitoring, this paper proposes a data-driven fault diagnosis methodology using the motor current signals of the ball screw. Two time-frequency domain analysis methods are investigated, including short-time Fourier transform (STFT) and wavelet packet decomposition (WPD). The statistical features are extracted, and Fisher score is used to select features. Furthermore, the logistic regression and k-nearest neighbors are applied for the final fault diagnosis. Experiments on a real-world industrial robot dataset are carried out for validation. 100% diagnosis accuracy can be basically achieved by the proposed method, which indicates the non-stationary current signal can be effectively used to identify the health states of the ball screw in the LCDTR.

Effective and reliable machinery health assessment and prognostic methods have been highly demanded in modern industries. In the past years, promising prognostic results have been achieved by the intelligent data-driven approaches. However, the existing methods generally rely on the availability of the complete system information. In the real industries, due to practical restrictions of mechanical structure, sensor installation etc ., the collected data may only cover partial system health information, which is denoted as partial observation problem. The existing data-driven methods are basically less effective in such scenarios with data incompleteness and disturbances. In order to address this issue, a deep learning-based remaining useful life prediction method is proposed in this article, where supervised attention mechanism is introduced. The informative data with more significant degradation features can be focused on for prognostics, and the data with nondiscriminative features can be ignored. A cutting wheel degradation image dataset is contributed to validate the proposed method, which is prepared from the real industrial manufacturing process. The experimental results suggest the proposed method offers an effective and promising approach on the prognostic problems with partial observations.

Extended from the spatially defined nearest neighbor index, the nearest neighbor index measures the levels of spatiotemporal clustering of a set of points, using only their spatial locations and th...

Quality inspection in semiconductor manufacturing is of great importance in the modern industries. In the recent years, intelligent data-driven condition monitoring methods have been successfully developed and applied in the industrial applications. However, despite the promising condition monitoring performance, the existing methods generally assume the training and testing data are from the same distribution. In practice, due to variations in manufacturing process, the collected data are usually subject to different distributions in different operating conditions, that significantly deteriorates the performance of the data-driven methods. To address this issue, this paper proposes a deep learning-based domain adaptation method for fault diagnosis in semiconductor manufacturing. The maximum mean discrepancy metric is optimized on the learned high-level data representation in the deep neural network. Experimental results on a real-world semiconductor manufacturing dataset suggest the proposed method offers an effective and generalized data-driven fault diagnosis approach for quality inspection.

Recently, the development of intelligent data-driven machinery fault diagnosis methods have received significant attention. In most studies, the training and testing data are assumed to be collected from the same sensor. However, in real practice, due to the mounting limitation and sensor malfunctioning, it cannot be generally guaranteed to obtain the data from the same sensor location at all times. The testing and training data can be possibly from different sensor locations. Consequently, different data distributions exist, which remarkably deteriorates the data-driven model performance in different scenarios. In order to address this issue, this paper proposes a deep learning-based cross-sensor domain adaptation approach for machinery fault diagnosis. The maximum mean discrepancy is deployed as a distance metric to realize marginal domain fusion. The unlabeled parallel data is further exploited to achieve conditional domain alignment with respect to different machine health conditions. An electro-mechanical actuator dataset is used as a case study for the validation of the proposed method. Different tasks are designed to simulate different cross-sensor domain adaptation problems in fault diagnosis. The experimental results suggest the proposed method achieves higher than $$95\%$$ testing accuracies in most tasks, and it offers a promising approach for cross-sensor fault diagnosis problems.

This study proposes a novel 1D deep convolutional transfer learning method that is able to learn the high-dimensional domain-invariant feature from the labeled training dataset and perform diagnosis tasks on the unlabeled testing dataset subjected to a domain shift. To obtain the domain-invariant features, the cross-entropy loss in the source domain classifier and the maximum mean discrepancies between the source and target domain data are minimized simultaneously. To evaluate the performance of the proposed method, an experimental study is conducted on a gearbox under significant speed variation. Because of inherent limitations of the vibration data, in this research, the effectiveness of torque measurement signals has been explored for gearbox fault diagnosis. Comprehensive studies on network parameters and the training sample size are performed to illustrate the robustness and effectiveness of the proposed method. A comparison study is performed on similar techniques to illustrate the superiority and high performance of the proposed diagnosis method. The achieved results illustrate the effectiveness of torque signal in multiclass cross-domain fault diagnosis of gearboxes.

As a promising modern technology, additive manufacturing (AM) has been receiving increasing research and industrial attention in the recent years. With its rapid development, the importance of quality monitoring in AM process has been recognized, which significantly affects the property of the manufactured parts. Since the conventional hand-crafted features for quality identification are generally costly, time-consuming and sensitive to noises, the intelligent data-driven automatic process monitoring methods are becoming more and more popular at present. This paper proposes a deep learning-based quality identification method for metal AM process. To alleviate the requirement for large amounts of high-quality labeled training data by most existing data-driven methods, an identification consistency-based approach is proposed to better explore the semi-supervised training data. The proposed method is able to achieve promising performance using limited supervised samples with low quality, such as noisy and blurred images. Experiments on a real-world metal AM dataset are implemented to validate the effectiveness of the proposed method, which offers a promising tool for real industrial applications.

Additive manufacturing (AM) techniques have been successfully developed in the past years with the great potential of overcoming the existing obstacles in traditional manufacturing. In order to improve the quality of the manufactured parts and reduce costs, it is important to timely and accurately monitor the AM process during manufacturing. However, it remains a challenging task due to the high complexity of the AM process and the difficulty in processing the condition monitoring data. This paper proposes a deep learning-based process monitoring method for directed energy deposition in AM. The thermal images collected during manufacturing are used to identify the process condition, and a deep convolutional neural network model is proposed to build an end-to-end condition monitoring framework. Experiments on a real directed energy deposition dataset in AM are carried out for validation. The results suggest the proposed method offers a promising approach in process monitoring based on the industrial images. Furthermore, little prior knowledge on signal processing and AM is required, that largely facilitates the potential applications in the real industrial scenarios.

In the recent years, the intelligent data-driven fault diagnosis methods on the rotating machines have been popularly developed. Especially, deep learning algorithms have been adopted in several studies and promising results have been obtained. However, the cross-domain fault diagnostic problem still remains a challenging issue, where the training and testing data are collected from different operating conditions of the machine. To bridge the domain gap in the training and testing data and increase the model generalization ability, a domain adaptation approach is proposed in this paper within the deep learning framework. Multiple convolutional operations are employed for automatic feature extraction. The maximum mean discrepancy is used to optimize the distributions of the learned features for accurate diagnosis. To further enhance the model generalization ability, the health condition labels of the data are injected with additional noise during model training. Experiments on a bearing fault diagnosis dataset are carried out for validation. The results show that, compared with the popular transfer learning algorithms, the proposed method with noisy health labels can successfully improve the cross-domain diagnosis performance on rotating machines.

Intelligent data-driven machinery health identification has been attracting increasing attention in the manufacturing industries, due to reduced maintenance cost and enhanced operation safety. Despite the successful development, the main limitation of most existing methods lies in the assumption that the training and testing data are collected from the same distribution, i.e. the same machine under identical condition. However, this assumption is difficult to be met in the real industries, since the diagnostic model is generally expected to be applied on new machines. In order to address this issue, a deep learning-based cross-machine health identification method is proposed for industrial vacuum pumps, which are of great importance in the manufacturing industry but have received far less research attention in the literature. Generalized diagnostic features can be learnt using the proposed domain adaptation technique with maximum mean discrepancy metric. The health identification model learnt from the training machines can be well applied on new machines. Experiments on a real-world vacuum pump dataset validate the proposed method, which is promising for industrial applications.

In Liquid Crystal Display (LCD) panel cutting, in-line monitoring of tool wear is important to maintain the dimension precision of finished products and to avoid possible damages to the workpiece. However, limited by the space for camera installation and the miniature structure of the tool itself, monitoring the wear of LCD panel cutting wheel is not a simple task. In this research, we proposed a machine-vision based instrumentation system and a systematic methodology for cutting wheel degradation monitoring. The proposed method describes the degradation of cutting wheel by estimating the tooth height of the cutting wheel blade based on the partially observed random samples. A series of methods are proposed to improve the reliability of the results. The effectiveness of the proposed method is validated based on the field data that is collected from three maintenance cycles. The validation results demonstrate consistent degradation trend of the cutting wheel over production cycles.

Difficulty in obtaining enough run-to-fail datasets is a major barrier that impedes the widespread acceptance of Prognostic and Health Management (PHM) technology in many applications. Recent progress in federated learning demonstrates great potential to overcome such difficulty because it allows one to train PHM models based on distributed databases without direct data sharing. Therefore, this technology can overcome local data scarcity challenges by training the PHM model based on multi-party databases. To demonstrate the ability of federated learning to enhance the robustness and reliability of PHM models, this paper proposes a novel federated Gaussian Mixture Model (GMM) algorithm to build universal baseline models based on distributed databases. A systematic methodology to perform collaborative prognostics is further presented using the proposed federated GMM algorithm. The usefulness and performance are validated through a simulated dataset and the NASA Turbofan Engine Dataset. The proposed federated approach with parameter sharing is shown to perform at par with the traditional approach with data sharing. The proposed model further demonstrates improved robustness of predictions made collaboratively keeping the data private compared to local predictions. Federated collaborative learning can serve as a catalyst for the adaptation of business models based on the servitization of assets in the era of Industry 4.0. The methodology facilitates effective learning of asset health conditions for data-scarce organizations by collaborating with other organizations preserving data privacy. This is most suitable for a servitization model for Overall Equipment Manufacturers who sell to multiple organizations.

Digital twin (DT) is gaining popularity due to its significant impacts on bridging the gap between the physical and cyber worlds. As reported by Grand View Research, Inc., the global market of DT is expected to reach $26.07 billion by 2025 with a Compound Annual Growth Rate of 38.2%. The growing adoption of cyber-physical system (CPS), Internet of Things, big data analytics, and cloud computing in manufacturing sector has paved the way for low cost and systematic implementation of DT, with promising impacts on (a) product design and development, (b) machine and equipment health monitoring, and (c) product support and services. Successful implementation of DT would increase transparency, cooperation, flexibility, resilience, production speed, scalability, and manufacturing efficiency. Realisation of smart manufacturing requires collaborative and autonomous interactions between sensing, networking, and computational resources across manufacturing assets where data is gathered from physical systems is utilised for the extraction of actionable insights and provision of predictive services. In this study, a reference architecture based on deep learning, DT, and 5C-CPS is proposed to facilitate the transformation towards smart manufacturing and Industry 4.0.

Modeling of dynamic manufacturing processes with slow drift using data-driven approaches is challenging because most data-driven models are trained by off-line data. In this paper, we propose to track slow drift of manufacturing processes using an online Bayesian Auto-Regression eXogenous (ARX) model with time-variant parameters. The model parameters are trained off-line and are updated online using Bayes’ rule. To avoid the frequent online model update, a Sample Importance (SI) test is proposed to screen online samples and only the sample that passes SI test is incremented to the VM model. The SI test decides the importance of the sample by collectively considering sample freshness, model prediction error and prediction uncertainty. Furthermore, the SI test is applied to off-line Data Base (DB) and an iteration algorithm is devised for off-line sample selection. The off-line sample selection algorithm can effectively reduce the sample redundancy in off-line DB while maintaining the model performance. To validate the effectiveness of the proposed method, Prognostics and Health Management (PHM) data challenge 2016 dataset is employed to predict material removal rate of chemical-mechanical planarization process. The validation results indicate that the proposed method outperforms existing state-of-art approaches in literature such as Just-in-Time (JIT) and deep learning models.

Machine learning with artificial neural network (ANN)-based methods is a powerful tool for the prediction and exploitation of the subtle relationships between the composition and properties of materials. This work utilizes an ANN to predict the composition of high-entropy alloys (HEAs) based on non-equimolar AlCoCrFeMnNi in order to achieve the highest hardness in the system. A simulated annealing algorithm is integrated with the ANN to optimize the composition. A bootstrap approach is adopted to quantify the uncertainty of the prediction. Without any guidance, the design of new compositions of AlCoCrFeMnNi-based HEAs would be difficult by empirical methods. This work successfully demonstrates that, by applying the machine learning method, new compositions of AlCoCrFeMnNi-based HEAs can be obtained, exhibiting hardness values higher than the best literature value for the same alloy system. The correlations between the predicted composition, hardness, and microstructure are also discussed.

Bottle capping is an important manufacturing process in many production systems. Bottle capping defects, if left unattended, will cause leakage issues during packaging and transportation. To develop a bottle capping defect detection and diagnosis system with lower cost, this paper proposes a data-driven methodology. Our approach is based on the motor current signal from the automatic capping machine, and four different capping defects, such as loose cap, cocked cap, and no cap, are considered in the experiments. Based on the current signal collected from the bottle capping machine, the signal envelope is extracted and smoothed first, and the faulty signatures are discovered subsequently. By designing and selecting appropriate features to describe the discovered fault signature, bottle capping defects can be effectively detected and isolated with the help of machine learning models. The results in the case study demonstrate that all the bottle capping defects considered in this investigation can be effectively detected and isolated.

As semiconductor manufacturing processes become more complex and sophisticated, manufacturers demand high quality, efficient process monitoring and fault detection (FD). Segmentation feature extraction (SFE) has been shown to be successful in FD when the anomaly patterns are well described by feature extraction, but not in cases of complex anomaly patterns. This paper proposes a novel approach that combines feature extraction-based and full trace analysis (FTA) capabilities in FD. First, a novel multivariate trace segmentation is applied to segment trace data based on defined patterns across key sensors, and then the pattern features and parameters are extracted within each segment. Second, full trace analysis generates adaptive guardbands on key sensors. These guardbands detect “anomaly” throughout the whole trace with a certain confidence interval. Third, a multivariate fault detection model is developed that combines SFE and FTA capabilities. Furthermore, multivariate data generation is developed to exhaustively validate combination and isolation of the proposed capabilities. Experimental results indicate that FTA can better capture the anomaly features with complex/atypical patterns, and the combined capability of SFE + FTA exhibits better performance than each method applied individually.

Research on scheduling problems is an evergreen challenge for industrial engineers. The growth of digital technologies opens the possibility to collect and analyze great amount of field data in real-time, representing a precious opportunity for an improved scheduling activity. Thus, scheduling under uncertain scenarios may benefit from the possibility to grasp the current operating conditions of the industrial equipment in real-time and take them into account when elaborating the best production schedules. To this end, the article proposes a proof-of-concept of a simheuristics framework for robust scheduling applied to a Flow Shop Scheduling Problem. The framework is composed of genetic algorithms for schedule optimization and discrete event simulation and is synchronized with the field through a Digital Twin (DT) that employs an Equipment Prognostics and Health Management (EPHM) module. The contribution of the EPHM module inside the DT-based framework is the real time computation of the failure probability of the equipment, with data-driven statistical models that take sensor data from the field as input. The viability of the framework is demonstrated in a flow shop application in a laboratory environment.

This paper presents an effective health assessment and predictive maintenance technique for industrial assets. The technique and algorithms applied to data sets provided by the Prognostics and Health Management Society 2014 Data Challenge. The data contains usage and part consumption for three years. In short, the usage data contains a parameter that roughly measures asset usage, and the part consumption data includes information regarding part replacement and maintenance actions. The first two years of data are considered as "Training" data and the third year is considered as "Testing" data. The proposed method built on the probability of the failure risk during training dataset. The main objective is to develop a model based on first two years data set (training) and determine the high risk and low risk times of failure for each individual asset for the third year.Training data shows many maintenance activities with 14 different codes. The principle difficulty is to detect the Preventive Maintenance (PM) in the training data. The paper presents the method in three main steps: the first step is to recognize the PM pattern based on time and type of maintenance activity via the training data. The second step is to determine the high-risk time intervals based on PM times by checking the frequency of the failures at specific times between each PM. The third step is to predict the high risk time intervals in the testing data using the information acquired from the training data. The score predicted by this probabilistic risk assessment method won the first place in the PHM Data Challenge Competition.

The neural network-based hyperspectral images (HSI) classification model has a deep structure, which leads to the increase of training parameters, long training time, and excessive computational cost. The deepened network models are likely to cause the problem of gradient disappearance, which limits further improvement for its classification accuracy. To this end, a residual unit with fewer training parameters were constructed by combining the residual connection with the depth-wise separable convolution. With the increased depth of the network, the number of output channels of each residual unit increases linearly with a small amplitude. The deepened network can continuously extract the spectral and spatial features while building a cone network structure by stacking the residual units. At the end of executing the model, a 1 × 1 convolution layer combined with a global average pooling layer can be used to replace the traditional fully connected layer to complete the classification with reduced parameters needed in the network. Experiments were conducted on three benchmark HSI datasets: Indian Pines, Pavia University, and Kennedy Space Center. The overall classification accuracy was 98.85%, 99.58%, and 99.96% respectively. Compared with other classification methods, the proposed network model guarantees a higher classification accuracy while spending less time on training and testing sample sites.

Human activity events are often recorded with their geographic locations and temporal stamps, which form spatial patterns of the events during individual time periods. Temporal attributes of these events help us understand the evolution of spatial processes over time. A challenge that researchers still face is that existing methods tend to treat all events as the same when evaluating the spatiotemporal pattern of events that have different properties. This article suggests a method for assessing the level of spatiotemporal clustering or spatiotemporal autocorrelation that may exist in a set of human activity events when they are associated with different categorical attributes. This method extends the Voronoi structure from 2D to 3D and integrates a sliding-window model as an approach to spatiotemporal tessellations of a space-time volume defined by a study area and time period. Furthermore, an index was developed to evaluate the partial spatiotemporal clustering level of one of the two event categories against the other category. The proposed method was applied to simulated data and a real-world dataset as a case study. Experimental results show that the method effectively measures the level of spatiotemporal clustering patterns among human activity events of multiple categories. The method can be applied to the analysis of large volumes of human activity events because of its computational efficiency.

In this article we explore Ni/HfO 2 /TiN resistive random access memory (RRAM) as ultralow power synaptic element in deep neural networks (DNN) for artificial intelligence applications. Low power RRAM devices are fabricated and measured, with very low RESET current and 2–3 orders of resistance window. The SET-RESET current-voltage characteristics, high- and low-resistance state statistical distribution, and analog programming characteristics are calibrated to analytical models. Training and inference for the MNIST handwritten digits dataset using a multilayer perceptron was simulated based on the calibrated model using CIMulator, a novel neuromorphic simulation platform for compute-in-memory circuitry to predict DNN inference accuracy and energy consumption. Despite larger inherent device-to-device variability due to low on current, 97% inference accuracy is achieved with only 2-bit for the weights using weight update accumulation technique.

The recent White House report on Artificial Intelligence (AI) (Lee, 2016) highlights the significance of AI and the necessity of a clear roadmap and strategic investment in this area. As AI emerges from science fiction to become the frontier of world-changing technologies, there is an urgent need for systematic development and implementation of AI to see its real impact in the next generation of industrial systems, namely Industry 4.0. Within the 5C architecture previously proposed in Lee et al. (2015), this paper provides an insight into the current state of AI technologies and the eco-system required to harness the power of AI in industrial applications.

Sparsity is becoming a more and more important topic in the area of machine learning and signal processing recently. One big family of sparse measures in current literature is the generalized l p / l q norm, which is scale invariant and is widely regarded as normalized l p norm. However, the characteristics of the generalized l p / l q norm are still less discussed and its application to the condition monitoring of rotating devices has been still unexplored. In this study, we firstly discuss the characteristics of the generalized l p / l q norm for sparse optimization and then propose a method of sparse filtering with the generalized l p / l q norm for the purpose of impulsive signature enhancement. Further driven by the trend of industrial big data and the need of reducing maintenance cost for industrial equipment, the proposed sparse filter is customized for vibration signal processing and also implemented on bearing and gearbox for the purpose of condition monitoring. Based on the results from the industrial implementations in this paper, the proposed method has been found to be a promising tool for impulsive feature enhancement, and the superiority of the proposed method over previous methods is also demonstrated.

Successful applications of Diffusion Map (DM) in machine failure detection and diagnosis have been reported in several recent studies. DM provides an efficient way to visualize the high-dimensional, complex and nonlinear machine data, and thus suggests more knowledge about the machine under monitoring. In this paper, a DM based methodology named as DM-EVD is proposed for machine degradation assessment, abnormality detection and diagnosis in an online fashion. Several limitations and challenges of using DM for machine health monitoring have been analyzed and addressed. Based on the proposed DM-EVD, a deviation based methodology is then proposed to include more dimension reduction methods. In this work, the incorporation of Laplacian Eigen-map and Principal Component Analysis (PCA) are explored, and the latter algorithm is named as PCA-Dev and is validated in the case study. To show the successful application of the proposed methodology, case studies from diverse fields are presented and investigated in this work. Improved results are reported by benchmarking with other machine learning algorithms.

Convolutional neural network (CNN) has shown excellent performance in hyperspectral image (HSI) classification. However, the structure of the CNN models is complex, requiring many training parameters and floating-point operations (FLOPs). This is often inefficient and results in longer training and testing time. In addition, the label samples of hyperspectral data are limited, and a deep network often causes the over-fitting phenomenon. Hence, a dual-path small convolution (DPSC) module is proposed. It is composed of two 1 × 1 small convolutions with a residual path and a density path. It can effectively extract abstract features from HSI. A dual-path small convolution network (DPSCN) is constructed by stacking DPSC modules. Specifically, the proposed model uses a DPSC module to complete the extraction of spectral and spectral–spatial features successively. It then uses a global average pooling layer at the end of the model to replace the conventional fully connected layer to complete the final classification. In the implemented study, all convolutional layers of the proposed network, except the middle layer, use 1 × 1 small convolution, effectively reduced model parameters and increased the speed of feature extraction processes. DPSCN was compared with several current state-of-the-art models. The results on three benchmark HSI data sets demonstrated that the proposed model is of lower complexity, has stronger generalization ability, and has higher classification efficiency.

Minimum Entropy Deconvolution (MED) filter, which is a non-parametric approach for impulsive signature detection, has been widely studied recently. Although the merits of the MED filter are manifold, this method tends to over highlight the dominant peaks and its performance becomes less stable when strong noise exists. In order to better understand the behavior of the MED filter, this study first investigated the mathematical fundamentals of the MED filter and then explained the reason why the MED filter tends to over highlight the dominant peaks. In order to pursue finer solutions for weak impulsive signature enhancement, the Convolutional Sparse Filter (CSF) is originally proposed in this work and the derivation of the CSF is presented in details. The superiority of the proposed CSF over the MED filter is validated by both simulated data and experimental data. The results demonstrate that CSF is an effective method for impulsive signature enhancement that could be applied in rotating machines for incipient fault detection.

In the rough machining stage, band saw machines are widely used to cut various raw materials into the required dimensions. The replacement of blade due to the blade degradation accounts for a large part of the total cost of band saw machine users. Therefore, mitigating the blade degradation by dynamically optimizing the cutting parameters can produce great economic benefits and is also a good exploration for smart manufacturing. To achieve this goal, the Convolutional Neural Network (CNN) model is proposed to map the complicated relationship between the cutting parameters and blade degradation. Then a simulation technique is used to search the optimal cutting parameters based on the blade degradation estimation from the model output, which would help alleviate the blade degradation. The proposed optimization method is validated on the data collected during the real manufacturing process of band saw machines. The comparative results demonstrate that the obtained optimal cutting parameters can effectively extend the service life of the blade for a band saw machine.

When abundant historical Run-to-Failure (R2F) data is available, the similarity-based method is one of the favored options for Remaining Useful Life (RUL) prediction due to its simplicity and satisfactory accuracy. In this study, a novel similarity-based methodology for RUL prediction is proposed. The proposed method has two important processing steps: similarity matching and Weibull fitting. The similarity matching screens the historical records by a similarity testing called Kernel Two Sample Test (KTST), and only those records that pass KTST are adopted as references for RUL prediction. For the selected similar records, the RUL is predicted as the remaining time to failure. The Weibull fitting fuses the multiple RUL predictions given by similar historical records. The PDF of RUL is estimated as the fitted Weibull distribution. To demonstrate the effectiveness and superiority of the proposed method, the famous CMAPSS data about aero-engine degradation is adopted for model validation. The results demonstrate improved prediction accuracy comparing with other similarity-based approaches and the state-of-the-art deep learning predictors.

For high-dimensional datasets, bad features and complex interactions between features can cause high computational costs and make outlier detection algorithms inefficient. Most feature selection methods are designed for supervised classification and regression, and limited works are specifically for unsupervised outlier detection. This paper proposes a novel isolation-based feature selection (IBFS) method for unsupervised outlier detection. It is based on the training process of isolation forest. When a point of a feature is used to split the data, the imbalanced distribution of split data is measured and used to quantify how strong this feature can detect outliers. We also compare the proposed method with variance, Laplacian score and kurtosis. These methods are benchmarked on simulated data to show their characteristics. Then we evaluate the performance using one-class support vector machine, isolation forest and local outlier factor on several real-word datasets. The evaluation results show that the proposed method can improve the performance of isolation forest, and its results are similar to and sometimes better than another useful outlier indicator: kurtosis, which demonstrate the effectiveness of the proposed method. We also notice that sometimes variance and Laplacian score has similar performance on the datasets.

Convolutional Neural Networks (CNN) are becoming increasingly popular for bearing fault diagnosis due to their ability to automatically capture the sensitive fault information without the need for expert knowledge. Most of these applications are developed considering vibration data from artificially induced faults. However, bearing failure in real-life can show huge damage variations even within a single category of failure which artificially induced failures are unable to represent. Thus, in this paper, the performance of classical CNN is evaluated on bearings with naturally occurring and progressing defects from the Paderborn University Dataset. A three-class (Healthy, Inner Race Fault and Outer Race Fault) classification problem is solved considering five bearing conditions within each class. These conditions vary in terms of bearing operating hours, damage mode, damage repetition pattern, the extent of damage, etc. The classification accuracy is evaluated under two cases: 1) at least a portion of data from each bearing condition from all classes is used in training; 2) data from all available conditions are considered for training except from one condition which is used explicitly for testing. Within each case, the effect of changing the domain of the input data is evaluated on the achieved accuracy. Three input signals based on vibration data (raw time domain signal, envelope spectrum, and spectrogram) were explored for their representation effectiveness. The proposed CNN with a spectrogram of the vibration signal as input achieves better results than similar architectures. Finally, the potential challenges that come along with the implementation of Deep Learning technologies for industrial applications are discussed and future research directions are proposed.

Gearboxes are widely used in rotating machinery and various industrial applications for transmission of power and torque. They operate for prolong hours and under different working conditions which may increase their probability of failure. Sudden failure of a gearbox may lead to significant downtime and increase maintenance costs. In industrial applications, usually fault detection and diagnosis techniques based on vibration signal are used for monitoring the health condition of gearboxes. In most of these techniques, time and frequency domain features are manually extracted from a vibration sensor and used for fault detection and diagnosis. In this research, a fault diagnosis methodology based on motor current signature analysis is proposed. The acquired data from multiple current sensors are fused by a novel 2-D convolutional neural network architecture and used for classification purpose directly without any need for manual feature extraction. Performance of the proposed method has been evaluated on the motor current data obtained from an industrial gearbox test rig in various health condition and with different working speeds. In comparison with classical machine learning (ML) algorithms, the presented methodology exhibits the best classification performance for gearbox fault detection and diagnosis.

Intelligent data-driven fault diagnosis methods have been successfully developed in the recent years. However, as one of the most important machines in the industries, the ball screw health monitoring problem has received less attention, due to the complex operating patterns and sophisticated mechanical structures. In practice, the working conditions of the ball screws usually change, that further makes the fault diagnosis problem more challenging since the data distributions are not the same. In order to address this issue, a deep learning-based domain adaptation method is proposed for the cross-domain ball screw fault diagnosis problem. The deep convolutional neural network is adopted for feature extraction and health condition classification. The maximum mean discrepancy metric is proposed to measure and optimize the data distributions of different operating conditions. A data segmentation method which is specially designed for the ball screw is further integrated. The experiments on the real ball screw condition monitoring data are carried out for validation. The results indicate the proposed approach is promising for the cross-domain diagnostic tasks of the ball screw in the real industries.

Existing intelligent gearbox fault diagnosis approaches have two shortcomings: (a) their performance is mostly confined to manual handcrafted features and (b) they follow a general assumption that the distribution of the data in source domain (labelled data on which the model is trained) is similar with target domain (unlabeled data on which model is tested); which might not be the case in real-world applications. Substantial human expertise and domain knowledge is required for manual feature extraction and moreover deploying the same model for a target domain whose distribution is different from the source domain would lead to poor generalization. Since deep learning methods can automatically learn high dimensional feature representations from raw measurement data, this paper proposes a novel deep learning based domain adaptation method for gearbox fault diagnosis under significant speed variations. A deep convolutional neural network is used as the main architecture. The paper proposes to minimize the summation of cross-entropy loss (between labelled source domain data) and Maximum Mean Discrepancy loss (between labelled source and unlabeled target datasets) simultaneously to adapt source domain model to be applied in the target domain. The proposed deep learning domain adaptation approach is evaluated using experimental data from a gearbox under variable speeds and multiple health conditions. An appropriate benchmarking with both traditional machine learning methods and other domain adaptation methods demonstrate the superiority of the proposed method.

This article aims to present a comprehensive review of the recent efforts and advances in applying machine learning (ML) techniques in the area of diagnostics and prognostics of rolling element bearings (REBs). The main goal of this study is to review, recognize and evaluate the performance of various ML techniques and compare them on criteria such as reliability, accuracy, robustness to noise, data volume requirements and implementation aspects. The merits and demerits of the reviewed ML techniques have been comprehensively analyzed and discussed. A comparative benchmarking of the performance of the reviewed ML algorithms is provided both from the viewpoint of theoretical aspects and industrial applicability. Finally, the potential challenges that come along with the implementation of ML technology are discussed in detail that will likely play a major role in the prognostics and health management of REBs. It is expected that this review will serve as a reference point for researchers to explore the opportunities for further improvement in the field of ML-based fault diagnosis and prognosis of REBs.

Spatial diffusion processes can be seen in many geographic phenomena that spread or migrate across space and over time. Studies of these processes were mostly done with verbal description until Hägerstrand (1966) started to approach it with quantitative models. A variety of attempts were made to continue this effort, but only with various degrees of success. Recognizing the critical role that distances between geographic objects or events play in a spatial diffusion process, we experimented with a new approach that uses these distances to detect and distinguish different types of spatial diffusion processes. Our approach is a two-step process that first calculates nearest neighbor ratios in a point process at each time step and then applies regression curve estimation to observe how these ratios change over time. We first report the results from applying this method to three spatio-temporal data sets which show the feasibility of our approach. We then report results of randomly simulated spatial diffusion processes to see if our approach is effective for the purpose of distinguishing different types of spatial diffusion processes. With only extreme cases as exceptions, our experiment found that using estimated regression curves of nearest neighbor ratios over time is usable in classifying spatial diffusion processes to either contagious/expansion or hierarchical/relocation diffusion processes.

Ant-based clustering is a type of clustering algorithm that imitates the behavior of ants. To improve the efficiency, increase the adaptability to non-Gaussian datasets and simplify the parameters of the algorithm, a novel ant-based clustering algorithm using Renyi Entropy (NAC-RE) is proposed. There are two aspects to application of Renyi entropy. Firstly, Kernel Entropy Component Analysis (KECA) is applied to modify the random projection of objects when the algorithm is run initially. This projection can create rough clusters and improve the algorithm's efficiency. Secondly, a novel ant movement model governed by Renyi entropy is proposed. The model takes each object as an ant. When the object (ant) moves to a new region, the Renyi entropy in its local neighborhood will be changed. The differential value of entropy governs whether the object should move or be moveless. The new model avoids complex parameters that have influence on the clustering results. The theoretical analysis has been conducted by kernel method to show that Renyi entropy metric is feasible and superior to distance metric. The novel algorithm was compared with other classic ones by several well-known benchmark datasets. The Friedman test with the corresponding Nemenyi test are applied to compare and conclude the algorithms' performance The results indicate that NAC-RE can get better results for non-linearly separable datasets while its parameters are simple.

Novelty detection is an important issue for practical industrial application, in which there is only normal operating data available in most cases. This paper proposes a systematic approach for novelty detection of mechanical components, using support vector data description (SVDD), a kernel approach for modeling the support of a distribution. To reduce the false alarm rate and increase the detection accuracy, a parameter optimization estimation scheme is proposed based on a grid search method that relies on the performance trade-off between the minimum fraction of support vectors and the maximum dual problem objective value. An evaluation value (E-value) chart based on the kernel distance for detection result is also designed to facilitate the decision visualization. To illustrate the effectiveness of the proposed method, novelty detection was applied to a particular kind of tapered roller bearing used in an industrial robot, which is investigated as a case study. The experimental results, in comparison to other methods, demonstrate that the proposed SVDD can conduct novelty detection of the monitored mechanical component effectively with higher accuracy.

Future industrial systems can be realized using the cyber–physical systems (CPSs) that advocate the coexistence of cyber and physical counterparts in a network structure to perform the system’s functions in a collaborative manner. Multiagent systems share common ground with CPSs and can empower them with a multitude of capabilities in their efforts to achieve complexity management, decentralization, intelligence, modularity, flexibility, robustness, adaptation, and responsiveness. This work surveys and analyzes the current state of the industrial application of agent technology in CPSs, and provides a vision on the way agents can effectively enable emerging CPS challenges. info:eu-repo/semantics/publishedVersion

WDFTC is a wavelet-based distribution-free CUSUM chart for detecting shifts in the mean of a profile with noisy components. Exploiting a discrete wavelet transform (DWT) of the mean in-control profile, WDFTC selects a reduced-dimension vector of the associated DWT components from which the mean in-control profile can be approximated with minimal weighted relative reconstruction error. Based on randomly sampled Phase I (in-control) profiles, the covariance matrix of the corresponding reduced-dimension DWT vectors is estimated using a matrix-regularisation method; then the DWT vectors are aggregated (batched) so that the non-overlapping batch means of the reduced-dimension DWT vectors have manageable covariances. To monitor shifts in the mean profile during Phase II operation, WDFTC computes a Hotelling's T 2-type statistic from successive non-overlapping batch means and applies a CUSUM procedure to those statistics, where the associated control limits are evaluated analytically from the Phase I data. Exper...

With the development of the condition-based maintenance techniques and the consequent requirement for good machine learning methods, new challenges arise in unsupervised learning. In the real-world situations, due to the relevant features that could exhibit the real machine condition are often unknown as priori, condition monitoring systems based on unimportant features, e.g. noise, might suffer high false-alarm rates, especially when the characteristics of failures are costly or difficult to learn. Therefore, it is important to select the most representative features for unsupervised learning in fault diagnostics. In this paper, a hybrid feature selection scheme (HFS) for unsupervised learning is proposed to improve the robustness and the accuracy of fault diagnostics. It provides a general framework of the feature selection based on significance evaluation and similarity measurement with respect to the multiple clustering solutions. The effectiveness of the proposed HFS method is demonstrated by a bearing fault diagnostics application and comparison with other features selection methods.

This paper proposes an online adaptive condition-based maintenance method with pattern discovery and fault learning capabilities for mechanical systems. The method is mainly based on a subtype of neural network techniques called self-organizing map (SOM). It is able to reduce local clusters from the same pattern and optimize the SOM architecture to further decrease the calculation cost in matching patterns in the neuron fitting process. Moreover, distance analysis and statistical pattern recognition (SPR) on neurons of the SOM are combined to establish rules and criteria for conducting and controlling the discovery and learning process so continuous process as purging prototypes on the map can be avoided. An experiment on condition monitoring of a machine tool test bed demonstrates and validates the effectiveness of the proposed approach.

PurposeThe purpose of this paper is to compare the effectiveness of different analytical approaches, namely artificial neural networks, logistic regression and support vector machines to assess the health of a strainer located at the suction side of the pump.Design/methodology/approachData used for simulation included exemplars from clean (represented by datasets after cleaning the suction strainer) and faulty conditions (represented by datasets prior to cleaning the suction strainer). The same datasets were used for modeling in order to compare how different techniques perform when fed with the same information.FindingsPrincipal component analysis‐based artificial neural networks proved to be better than other techniques in classifying maintenance datasets and predicting flow resistance from a clogged suction strainer.Originality/valueThe work highlights the comparative effectiveness of three predictive analytical techniques in classifying real plant data from a suction strainer. This will provide an opportunity for maintenance experts to see the effectiveness of different techniques as well as revealing valuable information about the relationship between the condition of the suction strainer and the overall performance of the pump.

In order to carry out precision and quality control of boring operations, on-line monitoring of boring tools is essential. Fourteen features were extracted by processing cutting force signals using virtual instrumentation. A Sequential Forward Search (SFS) algorithm was employed to select the best combination of features. Backpropagation neural networks (BPNs) and adaptive neuro-fuzzy inference systems (ANFIS) were used for on-line classification and measurement of tool wear. The input vectors consist of selected features. For the on-line classification, the outputs are boring tool conditions, which are either usable or worn out. For the on-line measurement, the outputs are estimated value of the tool wear. Using BPN, five features were needed for the on-line classification of boring tools. They are the average longitudinal force, average value of the ratio between the tangential and radial forces, skewness of the longitudinal force, skewness of the tangential force, and kurtosis of the longitudinal force. Three features, the average longitudinal force, average of the ratio between the tangential and radial forces, and kurtosis of the longitudinal force, were needed for on-line measurement of tool wear. Using ANFIS, three features were needed for the on-line classification of boring tools. They are the average longitudinal force, average of the ratio between the tangential and radial forces, and kurtosis of the longitudinal force. Only one feature, kurtosis of the longitudinal force, was needed for the on-line measurement of tool wear using ANFIS. Both 5x20x1BPN and 3x5 ANFIS can achieve a 100% success rate for the on-line classification of boring tool conditions. Using a 3x20x1BPN for neural computing, the minimum flank wear estimation error is 0.29% while the minimum flank wear estimation error is 2.04% using a 1x5 ANFIS.