Entropy, as a complexity measure, has been widely applied for time series analysis. One preeminent example is the design of machine condition monitoring and industrial fault diagnostic systems. The occurrence of failures in a machine will typically lead to non-linear characteristics in the measurements, caused by instantaneous variations, which can increase the complexity in the system response. Entropy measures are suitable to quantify such dynamic changes in the underlying process, distinguishing between different system conditions. However, notions of entropy are defined differently in various contexts (e.g., information theory and dynamical systems theory), which may confound researchers in the applied sciences. In this paper, we have systematically reviewed the theoretical development of some fundamental entropy measures and clarified the relations among them. Then, typical entropy-based applications of machine fault diagnostic systems are summarized. Further, insights into possible applications of the entropy measures are explained, as to where and how these measures can be useful towards future data-driven fault diagnosis methodologies. Finally, potential research trends in this area are discussed, with the intent of improving online entropy estimation and expanding its applicability to a wider range of intelligent fault diagnostic systems.

Condition monitoring and incipient fault diagnosis of rolling bearing is of great importance to detect failures and ensure reliable operations in rotating machinery. In this paper, a new multi-speed fault diagnostic approach is presented by using self-adaptive wavelet transform components generated from bearing vibration signals. The proposed approach is capable of discriminating signatures from four conditions of rolling bearing, i.e., normal bearing and three different types of defected bearings on outer race, inner race, and roller separately. Particle swarm optimization and Broyden-Fletche-Goldfarb-Shanno-based quasi-Newton minimization algorithms are applied to seek optimal parameters of Impulse Modeling-based continuous wavelet transform model. Then, a 3-D feature space of the statistical parameters and a nearest neighbor classifier are, respectively, applied for fault signature extraction and fault classification. Effectiveness of this approach is then evaluated, and the results have achieved an overall accuracy of 100%. Moreover, the generated discriminatory fault signatures are suitable for multi-speed fault data sets. This technique will be further implemented and tested in a real industrial environment.

The traditional means of monitoring the health of industrial systems involves the use of vibration and performance monitoring techniques amongst others. In these approaches, contact-type sensors, such as accelerometer, proximity probe, pressure transducer and temperature transducer, are installed on the machine to monitor its operational health parameters. However, these methods fall short when additional sensors cannot be installed on the machine due to cost, space constraint or sensor reliability concerns. On the other hand, the use of acoustic-based monitoring technique provides an improved alternative, as acoustic sensors (e.g., microphones) can be implemented quickly and cheaply in various scenarios and do not require physical contact with the machine. The collected acoustic signals contain relevant operating health information about the machine; yet they can be sensitive to background noise and changes in machine operating condition. These challenges are being addressed from the industrial applicability perspective for acoustic-based machine condition monitoring. This paper presents the development in methodology for acoustic-based fault diagnostic techniques and highlights the challenges encountered when analyzing sound for machine condition monitoring.

Recent advancements in smart sensors and deep learning facilitate the use of intelligent systems for machine health monitoring and diagnostics. While data-driven diagnosis methods can extract meaningful fault patterns automatically from sensor measurements, the reliability of such a bottom-up built system largely relies on the assumption – sensor readings are normal, without outliers and spurious readings. However, complex industrial environments or hardware malfunction is likely to cause noisy or corrupted sensor readings, resulting in degraded diagnosis performance. This paper proposes a multi-sensor based framework for fault diagnosis of rotating machinery, based on deep learning and data fusion techniques, integrating thermal imaging with vibration measurements. In contrast to the single-sensor method, the proposed method offers two advantages: improved robustness to background noise and diagnostic performance in analysing corrupted sensor readings. Three case studies are carried out to validate the effectiveness of the proposed method for multi-fault diagnosis of rotating machinery. The performance and trustworthiness of the system are studied and compared via analysing normal sensor data, data with different noise levels, and data with sensor anomaly (bias fault and stuck fault), respectively. The results demonstrate the proposed data fusion method presents a high diagnostic performance in identifying machine health conditions in a complex working environment.