Machinery Health Monitoring Research Articles

Self-supervised learning for tool wear monitoring with a disentangled-variational-autoencoder

The use of end-to-end deep learning in machinery health monitoring allows machine learning models to be created without the need for feature engineering. The research presented here expands on this use in the context of tool wear monitoring. A disentangled-variational-autoencoder, with a temporal convolutional neural network, is used to model and trend tool wear in a self-supervised manner, and anomaly detection is used to make predictions from both the input and latent spaces. The method achieves a precision-recall area-under-curve (PR-AUC) score of 0.45 across all cutting parameters on a milling dataset, and a top score of 0.80 for shallow depth cuts. The method achieves a top PR-AUC score of 0.41 on a real-world industrial CNC dataset, but the method does not generalise as well across the broad range of manufactured parts. The benefits of the approach, along with the drawbacks, are discussed in detail.

Coupled Hidden Markov Fusion of Multichannel Fast Spectral Coherence Features for Intelligent Fault Diagnosis of Rolling Element Bearings

Comprehensive understanding for machinery health monitoring could be realized by multisource information. Nevertheless, effective fusion of faulty features extracted from multichannel sensors remains challenging in intelligent fault diagnostics. For this reason, coupled hidden Markov model (CHMM) is proposed in this work to efficiently fuse spectral coherence (SC) features extracted from multichannel sensors for improving the diagnosis performance of rolling element bearings (REBs). To this end, SC features extracted from multichannel vibration data of the REB under different operational conditions are fused by a multichain CHMM. Parameter estimation algorithms and probabilistic inference for the CHMM are also developed in the addressed fusion approach. The effectiveness of the proposed method is validated by two different diagnosis experiments, one for fault classification and another for lifecycle performance evaluation of REBs. Compared with the state-of-the-art peer methods, the proposed multichannel feature fusion method has the best performance when dealing with fault diagnosis tasks for REBs.

Sensitivity comparison of impedance sensors with different coil arrangements

Two inductance-capacitance dual mode sensors based on different arrangement of dual coil and microchannel for oil multi-contaminant detection are proposed. The inductance detection model of metal particles and capacitance detection model of non-metal particles are constructed theoretically. The experimental results show that the two dual mode sensors both can realize inductance detection of iron and copper particles and capacitance detection of water droplets and air bubbles in oil. The detection sensitivity can be effectively improved by placing the microchannel on the edge of coil inner hole. The sensor of dual coils perpendicular to microchannel has more sensitive in inductance detection and the sensor of dual coils parallel to microchannel has more sensitive in capacitance detection. This research is helpful to improve the design and sensitivity of oil multi-contaminant detection sensor, and is of great significance to the realization of health monitoring and fault diagnosis of machinery.

Machinery Health Monitoring Based on Unsupervised Feature Learning via Generative Adversarial Networks

It confronts great difficulty to apply traditional artificial intelligence (AI) techniques to machinery prognostics and health management in manufacturing systems due to the lack of abnormal samples corresponding to different fault conditions. This article explores an unsupervised feature learning method for machinery health monitoring by proposing a generative adversarial networks (GAN) model that exploits the merits of the autoencoder and the traditional GAN. The major contribution is that the data distribution of the normal samples is accurately learned by the GAN model within both the signal spectrum and latent representation spaces. Specifically, the discriminative feature for machinery health monitoring is learned in an unsupervised manner by the proposed method in three steps. First, the proposed GAN model is trained by the normal samples of the inspected machine with the aim to correctly reconstruct the signal spectrum and its latent representation. Then, the trained model is applied to test the online samples of the same machine with unknown health conditions. Finally, the dissimilarity between the tested samples and their reconstructed ones in the latent representation space is taken as the discriminative feature. The feature value will increase significantly if a fault occurs in the inspected machine because the abnormal samples are never trained in the proposed GAN model. Experimental studies on three different machines are conducted to validate the proposed method and its superiority over the traditional methods in detecting abnormal points and characterizing fault propagation.

Bearing fault diagnosis using deep learning techniques coupled with handcrafted feature extraction: A comparative study

Deep learning has seen tremendous growth over the past decade. It has set new performance limits for a wide range of applications, including computer vision, speech recognition, and machinery health monitoring. With the abundance of instrumentation data and the availability of high computational power, deep learning continues to prove itself as an efficient tool for the extraction of micropatterns from machinery big data repositories. This study presents a comparative study for feature extraction capabilities using stacked autoencoders considering the use of expert domain knowledge. Case Western Reserve University bearing dataset was used for the study, and a classifier was trained and tested to extract and visualize features from 12 different failure classes. Based on the raw data preprocessing, four different deep neural network structures were studied. Results indicated that integrating domain knowledge with deep learning techniques improved feature extraction capabilities and reduced the deep neural networks size and computational requirements without the need for exhaustive deep neural networks architecture tuning and modification.

A Koopman operator approach for machinery health monitoring and prediction with noisy and low-dimensional industrial time series

Data-driven methods for machinery health monitoring and prediction, such as machine learning and statistical pattern recognition techniques, normally requires high quality (less noised), frequently-sampled and large volume time-series data to estimate proper models between system inputs and outputs. However, most of the industrial time-series are highly noisy and only few types of sensory data are available, which challenges the estimation accuracy. This paper proposes a data-driven spectral decomposition framework (denoted as Koopman-CBM) for the machinery health monitoring and prediction problem. Specifically, considering noisy industrial signals in the form of one dimensional time-series, we use the higher-order dynamic mode decomposition (DMD) embeds time-lagged snapshots to increase the spatial complexity of low-dimensional time series and use the total-least-square algorithm to compensate the effect of measurement noise, thereby, extracting accurate dynamical features. The obtained de-noised model characteristics (i.e., eigenvalues, eigenfunctions, Koopman operator) is effective in predicting the system health stages. In parallel, using the Koopman operator as the linear predictor associated with the nonlinear dynamics, we can then perform remaining useful life (RUL) predictions with high accuracy. The experimental validation of the proposed framework is carried out on a rolling bearing datasets for degradation health-stage inspection and RUL prediction. Results show that– compared with other mainstream methods– our approach is capable of identifying the critical degradation stages and achieving higher RUL prediction accuracies.

Generalized Vold–Kalman Filtering for Nonstationary Compound Faults Feature Extraction of Bearing and Gear

Effective detection of multifaults in bearings and gears is a challenging issue in rotary machinery health monitoring. As such, a generalized Vold–Kalman filtering (GVKF)-based compound faults diagnosis method is presented in this paper. The technique includes four main steps: 1) a time–frequency ridge is separated from the time–frequency representation (TFR) of the vibration signal using a peak search method; 2) according to the time–frequency ridge, GVKF parameters corresponding to all the fault characteristic frequencies (FCFs) are estimated; 3) the fault feature components are obtained using the generalized demodulation transform (GDT) and the VKF with the GVKF parameters; and 4) the spectra obtained by the fast Fourier transform (FFT) are used to fault detection. The main contributions of the proposed method are as follows: 1) the influence of speed fluctuations and the unrelated harmonic components are removed through the integration of the GDT and the VKF and 2) the tachometerless GVKF parameters are defined and calculated to quantitatively detect different fault types, which avoids missed diagnosis and misdiagnosis. The proposed multifault diagnosis algorithm is verified by both simulation and experiment data. Comparison with other commonly used techniques has shown the advantage of the new method.

Machine Learning Model for Event-Based Prognostics in Gas Circulator Condition Monitoring

Gas circulator (GC) units are an important rotating asset used in the advanced gas-cooled reactor design, facilitating the flow of CO $_2$ gas through the reactor core. The ongoing maintenance and examination of these machines are important for operators in order to maintain safe and economic generation. GCs experience a dynamic duty cycle with periods of nonsteady state behavior at regular refueling intervals, posing a unique analysis problem for reliability engineers. In line with the increased data volumes and sophistication of available technologies, the investigation of predictive and prognostic measurements has become a central interest in rotating asset condition monitoring. However, many of the state-of-the-art approaches finding success deal with the extrapolation of stationary time series feeds, with little to no consideration of more complex but expected events in the data. In this paper, we demonstrate a novel modeling approach for examining refueling behaviors in GCs, with a focus on estimating their health state from vibration data. A machine learning model was constructed using the operational history of a unit experiencing an eventual inspection-based failure. This new approach to examining GC condition is shown to correspond well with explicit remaining useful life measurements of the case study, improving on the existing rudimentary extrapolation methods often employed in rotating machinery health monitoring.

A Deep Learning-based Method for Machinery Health Monitoring with Big Data

A Deep Learning-based Method for Machinery Health Monitoring with Big Data

Machinery Fault Diagnosis Using Two-Channel Analysis Method Based on Fictitious System Frequency Response Function

Most existing techniques for machinery health monitoring that utilize measured vibration signals usually require measurement points to be as close as possible to the expected fault components of interest. This is particularly important for implementing condition-based maintenance since the incipient fault signal power may be too small to be detected if a sensor is located further away from the fault source. However, a measurement sensor is often not attached to the ideal point due to geometric or environmental restrictions. In such a case, many of the conventional diagnostic techniques may not be successfully applicable. In this paper, a two-channel analysis method is proposed to overcome such difficulty. It uses two vibration signals simultaneously measured at arbitrary points in a machine. The proposed method is described theoretically by introducing a fictitious system frequency response function. It is then verified experimentally for bearing fault detection. The results show that the suggested method may be a good alternative when ideal points for measurement sensors are not readily available.

Structural Health Monitoring of Rotating Machines in Manufacturing Processes by Vibration Methods

The paper presents results of the active diagnostics experiments on influence of fatigue metal damage of the inner race of bearing and unbalance of rotating masses on vibration generated by the machine. Analysis of vibration related phenomena is a solution commonly applied in Structural Health Monitoring (SHM) systems. The application of vibroacoustics methods for technical condition monitoring has been developed in the past years in many systems of manufacturing processes. Vibroacoustic methods, based on the analysis of vibration or acoustic signals perceived as residual processes of non-invasive nature, is becoming more and more important in this respect. The scope of its application as well as the applicability of methods in numerous diagnostic systems also results from the capabilities of advanced methods of signal analysis and identification of numerous characteristics of technical condition. One of the most common operation damages are caused by rolling bearings wear. The scope of research contains tests on bearing damage and the unbalance of disc. The wear processes and unbalance are closely related to the vibration levels (arising from the mass loss of plastic deformation, and the fatigue damage). The research was conducted on special research test bench for vibration monitoring for rotating machinery. Structural health monitoring of machinery has to be conducted in different states and working conditions of the manufacturing system. Thus for simulating of different operating conditions the experiments have been conducted during run up of the machine which consist the transient states of working and during work on constant rotational speed of the power generate engine. For the identification of the symptoms of the machinery and equipments health monitoring the vibration signal have been analysed in time domain and frequency transformation as well. The performed signals are not stationary. Thus it is better to observe the signal simultaneously in time and frequency domains. For this purpose the spectrograms were determined. Spectrograms computes the short-time Fourier transform of a signal by default divided into segments. During the transformation the Hamming window and noverlap were used. For the comparison of the vibration of good and damage bearings signals registered for different frequencies have been presented in form of spectrograms and RMS distributions.

Self-Diagnostics of Piezoelectric Transducers

Equipment technical state monitoring systems contain hundreds of vibration and temperature sensors for real-time health monitoring of machinery. Monitoring systems reliability should significantly exceed the units stock reliability, being protected from accidents and incidents.The paper considers the approach to monitoring systems reliability growth, primarily by means of providing control (monitoring) of measuring-converter path technical condition. Sensors are the weakest point, due not only to wear, but also to damage, caused by repair personnel when servicing equipment. The self-diagnostics method for channels with piezoelectric transducers and the choice of optimal self-diagnostics circuits parameters have been suggested. The results of practical implementation have been provided as well.

Time series trending for condition assessment and prognostics

Purpose – The developments of complex systems have increased the demand for condition monitoring techniques so as to maximize operational availability and safety while decreasing the costs. Signal analysis is one of the methods used to develop condition monitoring in order to extract important information contained in the sensory signals, which can be used for health assessment. However, extraction of such information from collected data in a practical working environment is always a great challenge as sensory signals are usually multi-dimensional and obscured by noise. The paper aims to discuss this issue. Design/methodology/approach – This paper presents a method for trends extraction from multi-dimensional sensory data, which are then used for machinery health monitoring and maintenance needs. The proposed method is based on extracting successive features from machinery sensory signals. Then, unsupervised feature selection on the features domain is applied without making any assumptions concerning the source of the signals and the number of the extracted features. Finally, empirical mode decomposition (EMD) algorithm is applied on the projected features with the purpose of following the evolution of data in a compact representation over time. Findings – The method is demonstrated on accelerated degradation data set of bearings acquired from PRONOSTIA experimental platform and a second data set acquired form NASA repository. Originality/value – The method showed that it is able to extract interesting signal trends which can be used for health monitoring and remaining useful life prediction.

Analysis of Vibration in gearbox sensor data using Lipschitz Exponent (LE) function: A Wavelet approach

Machinery health monitoring is a key step in the implementation of maintenance in industry. A remarkable property of the wavelet transform is its ability to characterize the local regularity of machines. The Lipschitz exponent (LE) is the most popular measure of the regularity behavior of a signal. In this paper we adopt a new area-based objective function used to determine the Lipschitz exponent value. We analyzed four different set of gear box data such as normal, slight ward, medium warm and one tooth broken condition. We applied statistical features are commonly used to provide a measure of the vibration level that can be compared to a threshold value indicative of a failed condition. Through the analysis and diagnosis of measuring vibration signal using LE based objective function, the algorithm is verified to be reliable and effective.

Unsupervised Trend Extraction for Prognostics and Condition Assessment

Maintenance is becoming more expensive nowadays due to the increased complexity in design and function of industrial systems. Continuous health monitoring is thus of high importance to increase the availability of industrial systems and consequently reduce the costs. This paper presents an algorithm for unsupervised trends extraction from multidimensional sensory data so as to use such trend in machinery health monitoring and maintenance needs. The proposed method does not assume any prior knowledge about the nature and type of the input signals. It is based on extracting successive multi-dimensional features from machinery sensory signals. Then, unsupervised feature selection on the features domain is applied without making any assumptions concerning the source of the signals and the number of the extracted features. Finally, empirical mode decomposition algorithm (EMD) is applied on the projected features with the purpose of following the evolution of data in a compact representation over time. The algorithm is demonstrated on accelerated degradation dataset of bearings acquired from PRONOSTIA experimental platform and a second dataset acquired form NASA repository where it is shown to be able to extract interesting signal trends.

New technology of subsea and offshore direct drive compressor with magnetic bearing

Subsea compressor offers tremendous potentials, but also some obstacles. Using magnetic bearings can eliminate lubrication oil system as well as seal system. High-speed permanent-magnet electrical-motor direct-drive technology allows centrifugal compressor to meet optimum speed, eliminating gear box. This design offers inherent machinery health monitoring features and very compact and reliable train. New technology of hermetically sealed direct drive centrifugal compressor for offshore and subsea applications are reviewed and discussed. Latest technologies and modern achievements on direct drive compressor designs, magnetic bearings, permanent magnet motors, cooling systems and train arrangements are reviewed. Case studies for horizontal and vertical compressor trains are presented.

Structural Health Management in the NAVY

There is a critical need for integrated system health management (ISHM) approaches to asset maintenance. Ideally, ISHM methodologies would track the system usage and the associated loads, monitor the system degradation and materials state, monitor relevant environmental parameters and their effects on system degradation, detect insipient system damage, diagnose failure mode, predict future system performance, and recommend maintenance actions. Even though there has been considerable progress in many subareas of ISHM over the past years, there is still ample room for future improvements in all technological aspects affecting ISHM. In fact, progress in ISHM has not been uniform. Some subsystems have experienced a far greater degree of development than others. For example, engine and machinery health monitoring and diagnostics, due to its criticality, has evolved at a faster pace than structural health monitoring. This article will review some of the aspects that need to be addressed in order to make structural health monitoring (SHM) of military systems a reality in the near future.

The application of fiber grating sensor network in port-machinery monitoring

A Fiber Bragg Grating sensor network is designed to implement the real-time health monitoring of port machinery. This paper describes the system design and its application in port-machinery monitoring. The results of field test on port machinery in Qing Huangdao (China) are discussed. Through the long-term monitoring on running state, especially the structure strain of port machinery, the status of the port machinery including security and fatigue life could be acquired and estimated exactly. It could provide the foundation for the use and maintenance of port machinery, and the security of the port machinery is improved. Long-term test results show good stability and repeatability of <8 με over two years operation with a precision of ±4 με and a resolution of 1 με.

An optical system with potential for remote health monitoring of subsea machinery

A prototype fibre-optic sensing system is described with potential to remotely monitor the condition of three-phase variable frequency subsea motors and electric submersible pumps. An indication that the integrity of a powerful electric motor may be compromised can be gained by spectral analysis of the stator drive current, the phases of the currents, the measurement of vibration at specific locations on the motor and the temperature of the bearings. The optical interrogation system is based on an imbalanced Mach–Zehnder fibre interferometer, illuminated with a broadband source with fibre Bragg gratings (FBGs) used as the basic transducers for the current and vibration measurements. Signals were recovered at a distance of 7 km, for the ‘in house’ current sensor at an effective ac current of 500 A at frequencies from 10 to 1000 Hz with a S/N ratio of 45 dB, together with the relative phases between the currents. These signals were generated with a simulated high power three-phase electric motor. A commercial accelerometer was incorporated into the system which demonstrated a sensitivity of 1 mg Hz−1/2 also from 10 to 1000 Hz. As the interrogation system can only be used to detect dynamic signals, a second interrogation for quasi-static temperature measurements is required; although this was not deployed, possible applications other than bearing temperature measurement are considered.

Singularity detection in machinery health monitoring using Lipschitz exponent function

Machinery health monitoring is a key step in the implementation of Condition-based Maintenance in industry. In this procedure, a quantitative description of machine health condition is necessary for maintenance decision-making. In this paper, we applied singularity analysis with wavelet for data processing and a new concept, Lipschitz exponent function, was proposed based on wavelet transform. A kurtosis based health index was defined, which can be used for maintenance decision-making. The proposed method was validated with two sets of gearbox vibration data in comparison with three other indexes. The results show that kurtosis based health index demonstrates excellent performance.