Search

Tool condition monitoring (TCM) is crucial to ensure good quality products and avoid downtime. Machine learning has proven to be vital for TCM. However, existing works are predominately based on supervised learning, which hinders their applicability in real-world manufacturing settings, where data labeling is cumbersome and costly with in-service machines. Additionally, the existing unsupervised solutions mostly handle binary decision-based TCM which is unable to fully reflect the dynamics of tool wear progression. To address these issues, we propose different unsupervised and semisupervised five-class tool wear recognition frameworks to handle fully unlabeled and partially labeled data, respectively. The underlying methods include Laplacian score, sparse autoencoder (SAE), stacked SAE (SSAE), self-organizing map, Softmax, support vector machine, and random forest. For the semisupervised frameworks, we considered designs where labeled data influence only feature learning, classifier building, or both. We also investigated different training configurations of SSAE regarding the supervision level. We applied the frameworks on two run-to-failure datasets of milling tools, recorded using a microphone and an accelerometer. Single sensor and multisensor data under different percentages of labeled training data were considered in the evaluation. The results showed which of the frameworks led to the best predictive performance under which data settings, and highlighted the significance of sensor fusion and discriminative feature representations in combating the unavailability and scarcity of labels, among other findings. The highest macro-F1 achieved for the two datasets with fully unlabeled data reached 87.52% and 75.80%, respectively, and over 90% when only 25% of the training observations were labeled.

Predictive maintenance (PdM) has become an important industrial feature. Existing methods mainly focus on remaining useful life (RUL) regression or anomaly detection to achieve PdM in a given application. Those approaches assume monotonic degradation processes leading to a single catastrophic failure at the system's end of lifetime. In contrast, much more complex degradation processes can be found in real-world applications, which are characterized by effects like self-healing or noncatastrophic anomalies. A important example of devices with complex degradation are electromechanical relays. As established PdM solutions failed when applied to a real-world relays degradation data set, the maintenance algorithm for unlabeled data (MAUD) is presented to detect signs of wear and enable a service in time. In detail, MAUD is based on an artificial neural network (ANN), which is trained semisupervised. Experiments with measurement data from 546 relays show that MAUD is superior to various existing methods: The static B10 threshold, which represents the state of the art in relay maintenance, is surpassed by a 17.07 p.p. increase in utilization while reducing failures by 6.42 p.p. Methods based on machine learning, such as RUL estimation and anomaly detection, achieved much lower utilization (up to 31.83 p.p.) compared with MAUD while maintaining the same failure rate.

Predictive maintenance (PdM) uses statistical and machine learning methods to detect and predict the onset of faults. PdM is often used in industrial IoT settings in the energy sector, where research works usually consider specific types of faults depending on the application. However, since PdM is mainly data-driven and needs to work in real time, the public availability of datasets is required in order to build efficient and effective models applicable across multiple domains. Unlike methods, the publicly available datasets obtained from sensors in the energy sector have not been properly reviewed or categorized. In this work, we consider five subsectors of the energy sector: wind, solar, oil & gas, diesel & thermal, and electrical power grid. We provide a detailed description of the properties of the publicly available PdM datasets in these subsectors. The review of the datasets is conducted on a number of scientific and commercial repositories: IEEE DataPort, UCI Machine Learning Repository, Kaggle, EDP, and Mendeley Data. The datasets are graded into three categories according to objective criteria. We also provide references to significant related research work that uses the considered datasets. The observed challenges in using the datasets in this field are thoroughly discussed. We find that there is a troublesome scarcity of publicly available datasets in the energy sector, more so of data coming from real, non-simulated sources. Three datasets, 3W (oil & gas), EDP-WT (wind), and OREC (wind) stand out as highly valuable for researchers in this field.

In a semiconductor plasma etcher, it is becoming increasingly necessary to improve productivity by reducing unplanned equipment maintenance. Thus, predictive maintenance (PdM) is typically conducted using equipment data to predict the failure timing, after which proactive measures should be taken. In PdM, the planned maintenance schedule is updated on the basis of the predicted failure timing. However, in practice, the predicted failure timing has a probabilistic variability. Therefore, we propose a maintenance schedule update method based on the expected maintenance cost calculated from the probabilistic variability of the failure timing. We applied our method and conventional methods to a dataset of failure cases that model actual component failures of etchers and found that our method was effective in terms of reducing maintenance costs. [ABSTRACT FROM AUTHOR]

In this article, we present a low-power (LP) anomaly detection integrated circuit (ADIC) based on a one-class classifier (OCC) neural network. The ADIC achieves LP operation through a combination of: 1) careful choice of algorithm for online learning and 2) approximate computing techniques to lower average energy. In particular, online pseudoinverse update method (OPIUM) is used to train a randomized neural network for quick and resource-efficient learning. An additional 42% energy saving can be achieved when a lighter version of OPIUM method is used for training with the same number of data samples lead to no significant compromise on the quality of inference. Instead of a single classifier with large number of neurons, an ensemble of $K$ base learner (BL) approach is chosen to reduce learning memory by a factor of $K$. This also enables approximate computing by dynamically varying the neural network size based on anomaly detection. Fabricated in 65-nm CMOS, the ADIC has $K=7$ BLs with 32 neurons in each BL and dissipates 11.87 and 3.35 pJ/OP during learning and inference, respectively, at $V_{\text {dd}}=0.75\,\,\text {V}$ when all seven BLs are enabled. Furthermore, evaluated on the NASA bearing data set, approximately 80% of the chip can be shut down for 99% of the lifetime leading to an energy efficiency of 0.48 pJ/OP, an $18.5 \times $ reduction over full-precision computing running at $V_{\text {dd}}=1.2 \,\, \text {V}$ throughout the lifetime. [ABSTRACT FROM AUTHOR]

Artificial intelligence, big data, machine learning, cloud computing, and Internet of Things (IoT) are terms which have driven the fourth industrial revolution. The digital revolution has transformed the manufacturing industry into smart manufacturing through the development of intelligent systems. In this paper, a big data ecosystem is presented for the implementation of fault detection and diagnosis in predictive maintenance with real industrial big data gathered directly from large-scale global manufacturing plants, aiming to provide a complete architecture which could be used in industrial IoT-based smart manufacturing in an industrial 4.0 system. The proposed architecture overcomes multiple challenges including big data ingestion, integration, transformation, storage, analytics, and visualization in a real-time environment using various technologies such as the data lake, NoSQL database, Apache Spark, Apache Drill, Apache Hive, OPC Collector, and other techniques. Transformation protocols, authentication, and data encryption methods are also utilized to address data and network security issues. A MapReduce-based distributed PCA model is designed for fault detection and diagnosis. In a large-scale manufacturing system, not all kinds of failure data are accessible, and the absence of labels precludes all the supervised methods in the predictive phase. Furthermore, the proposed framework takes advantage of some of the characteristics of PCA such as its ease of implementation on Spark, its simple algorithmic structure, and its real-time processing ability. All these elements are essential for smart manufacturing in the evolution to Industry 4.0. The proposed detection system has been implemented into the real-time industrial production system in a cooperated company, running for several years, and the results successfully provide an alarm warning several days before the fault happens. A test case involving several outages in 2014 is reported and analyzed in detail during the experiment section. [ABSTRACT FROM AUTHOR]

With the tremendous revival of artificial intelligence, predictive maintenance (PdM) based on data-driven methods has become the most effective solution to address smart manufacturing and industrial big data, especially for performing health perception (e.g., fault diagnosis and remaining life assessment). Moreover, because the existing PdM research is still in primary experimental stage, most works are conducted utilizing several open-datasets, and the combination with specific applications such as rotating machinery is especially rare. Hence, in this paper, we focus on data-driven methods for PdM, present a comprehensive survey on its applications, and attempt to provide graduate students, companies, and institutions with the preliminary understanding of the existing works recently published. Specifically, we first briefly introduce the PdM approach, illustrate our PdM scheme for automatic washing equipment, and demonstrate the challenges encountered when we conduct a PdM research. Second, we classify the specific industrial applications based on six algorithms of machine learning and deep learning (DL), and compare five performance metrics for each classification. Furthermore, the accuracy (a metric to evaluate the algorithm performance) of these PdM applications is analyzed in detail. There are some important conclusions: 1) the data used in the summarized literature are mostly from public datasets, such as case western reserve university (CWRU)/intelligent maintenance systems (IMS); and 2) in recent years, researchers seem to focus more on DL algorithms for PdM research. Finally, we summarize the common features regarding our surveyed PdM applications and discuss several potential directions. [ABSTRACT FROM AUTHOR]

Considering the operation process rebuilding (OPR) of manufacturing/operation systems, we propose a dynamic interactive bilevel maintenance methodology to satisfy rapid market changes. Predictive maintenance (PdM) intervals at the machine level are dynamically scheduled by a multiobjective model for each diverse machine. A system-level opportunistic maintenance (OM) policy is proposed to facilitate PdM optimizations according to OPR activities. This novel OPR-OM policy utilizes a variable maintenance time window to construct optimal maintenance schedules that are suitable for changeable system structures. The results obtained by applying this methodology at Shanghai Port indicate that the proposed methodology can help a port transportation system to achieve rapid responses to OPR activities, which can significantly improve system efficiency and economy. [ABSTRACT FROM AUTHOR]

Most conventional FDC approaches are used to find the TDs required for monitoring and the TDs' related key parameters that need to be monitored, and then apply the SPC approach to detect the faults. However, in a practical situation, an abnormal key-parameter value may not be caused solely by its own TD; it may result from the other related parameters. Therefore, accurate fault classification or diagnosis may not be achieved. Moreover, most conventional PdM methods require a library of degradation patterns from previous run-to-failure data sets. Without those massive historical failure data, the PdM methods may not function properly. In this paper, we propose a virtual-metrology- (VM) based BPM scheme that possesses the capabilities of FDC and PdM. The BPM scheme contains the TD baseline model, FDC logic, and a RUL predictive module. The TD baseline model generated by the VM technique is applied to serve as the reference for detecting the fault. By applying the BPM scheme, fault diagnosis and prognosis can be accomplished, the problem of the conventional SPC method mentioned above can be resolved, and the requirement of massive historical failure data can also be released. [ABSTRACT FROM PUBLISHER]

Silicon epitaxial deposition is a process strongly influenced by wafer temperature behavior, which has to be constantly monitored to avoid the production of defective wafers. However, temperature measurements are not reliable, and the sensors have to be appropriately calibrated with some dedicated procedure. A predictive maintenance (PdM) system is proposed with the aim of predicting process behavior and scheduling control actions on the sensors in advance. Two different prediction techniques have been employed and compared: the Kalman predictor and the particle filter with Gaussian kernel density estimator. The accuracy of the PdM module has been tested on real industrial production datasets. [ABSTRACT FROM AUTHOR]