Subscribe to download publications


Emphasizing bearing tones for prognostics


Bearing failure is one of the foremost causes of breakdowns in rotating machinery and such failure can be catastrophic, resulting in costly downtime. One of the key issues in bearing prognostics is to detect the defect at its incipient stage and alert the operator before it develops into a catastrophic failure.

Vibration analysis can be applied very effectively to rolling element bearing condition monitoring in all three aspects (detection, diagnosis, and prognosis).

Signal de-noising and extraction of the weak signature are crucial to bearing prognostics since the signature of a defective bearing is spread across a wide frequency band and, hence, can easily become masked by other sources of excitation and noise. One of the challenges is to enhance the weak signature at the early stage of defect development.

The proposed method separates bearing signals from signals of shafts, gears, rotors, pumps, etc. (in fact all the discrete frequencies) since the latter usually dominate the spectrum. The technique is based on a few stages of resampling and removal of the synchronous time average of the vibration signal. The results show the effectiveness of the method for diagnosis and for automatic bearing feature extraction both in terms of orders in the spectrum and orders in the envelope spectrum representation.

Environment for processing of wideband signals


Acquisition and analysis of signals from mechanical equipment are necessary steps to achieve the goal of fully automatic diagnostics and prognostics. When some of the data is wide-band, such as signals from vibration and acoustic sensors, the processing stage is computationally intensive and requires a sophisticated handling environment.

The article presents a proposed architecture for such an environment. This architecture was developed and used successfully in R.K. Diagnostics, and is being offered to its customers. The environment is able to support multiple mechanical systems, different configurations of acquisition equipment, automatic data screening, automatic recognition of operation modes, as well as facilitating flexible flows of algorithms configurable for different combinations of flight regimes, plants, or sensors.

Other features of the architecture include providing a simple interface for development of diagnostics and prognostics algorithms, ability to store most of the parameters in external database, and ability to export algorithms to customer embedded platform. The system is OSA/CBM compliant, highly modular, platform independent and flexible.

Analysis and illustration of the proposed environment is presented for an application of vibrations data analysis using MATLAB.

Model Based Approach for Identification of Gears and Bearings Failure Modes


This paper describes the algorithms that were used for analysis of the PHM’09 gear-box. The purpose of the analysis was to detect and identify faults in various components of the gear-box. Each of the 560 vibration recordings presented a different set of faults, including distributed and localized gear faults, typical bearing faults and shaft faults. Each fault had to be pinpointed precisely.

In the following sections we describe the algorithms used for finding faults in bearings, gears and shafts, and the conclusions that were reached. A special blend of pattern recognition and signal processing methods was applied.

Bearings were analyzed using the orders representation of the envelope of a band pass filtered signal and an envelope of the de-phased signal. A special search algorithm was applied for bearing feature extraction. The diagnostics of the bearing failure modes was carried out automatically. Gears were analyzed using the order domains, the quefrency of orders, and the derivatives of the phase average.

Methods for diagnostics of bearings in non-stationary environment


Bearing failure is one of the major causes of breakdown in rotating machinery. One of the key challenges in bearing diagnostics and prognostics is to detect the defect as early as possible, when the failure signs are weak.

Vibration based early detection of bearing failure requires improvement of the signal to noise ratio by an effective signal de-noising and extraction of the weak failure signs that can be obscured by other vibration sources and noise. The challenge is to enhance the weak signature in the early stages of defect development.

The task of enhancing the weak failure signs is complicated by the fact that changes in operating conditions influence vibrations sources and change the mixture recorded by the sensors. As a result, the recorded signal becomes non-stationary. The proposed technique suggests a solution to that problem. The technique adapts the basic dephase method to analysis of non-stationary signals recorded from a system operating under changing operating conditions.

The adapted dephase method was applied to vibrations measured in rotating machinery, including systems with healthy bearings and damaged bearings. The same feature extraction procedure was applied to signatures in the orders domain before and after the application of the adapted dephase. The results show the effectiveness of the method for diagnosis both in the orders representation and in the orders of the envelope representation.

Decision and Fusion for Diagnostics of Mechanical Components


Detection of damaged mechanical components in their early stages is crucial in many applications. The diagnostics of mechanical components is achieved most effectively using vibration and/or acoustical measurements, sometimes accompanied by oil debris indications.

The paper describes a concept for fusion and decision for mechanical components, based on vibro-acoustic signatures. Typically in diagnostics of complex machinery, there are numerous records from normally operating machines and few recordings with damaged components. Diagnostics of each mechanical component requires consideration of a large number of features.

Learning classification algorithms cannot be applied due to insufficient examples of damaged components. The proposed system presents a solution by introducing a hierarchical decision scheme. The proposed architecture is designed in layers imitating expert’s decision reasoning. The architecture and tools used allow incorporation of expert’s knowledge along with the ability to learn from examples. The system was implemented and tested on simulated data and real-world data from seeded tests. The paper describes the proposed architecture, the algorithms used to implement it and some examples.

Bearing diagnostics using image processing methods


In complex machines, the failure signs of an early bearing damage are weak compared to other sources of excitations (e.g. gears, shafts, rotors, etc.). The task of emphasizing the failure signs is complicated by the fact that changes in operating conditions influence vibrations sources and change the frequency and amplitude characteristics of the signal, making it non-stationary. As a result, a joint time-frequency representation is required. Previous vibration based diagnostic techniques focused on either the time domain or the frequency domain.

The proposed method suggests a different solution that applies image processing techniques to time-frequency or RPM-order representations (TFR) of the vibration signals in the orders-RPM domain.

In the first stage, TFRs of healthy machines are used to create a baseline. The TFRs can be obtained using various methods (Wigner-Ville, wavelets, STFT, etc).

In the next stage, the distance TFR between the inspected recording and the baseline is computed. In the third stage, the distance TFR is analyzed using ridge tracking and other image processing algorithms. In the fourth stage, the relations between the detected ridges are compared to the characteristic patterns of the bearing failure modes and the matching ridges are selected.

The different stages of analysis: baselines, distance TFR, ridges detection and selection, are illustrated with actual data of damaged bearings.

Methods for bearing feature extraction


Vibration based detection of bearing failure requires careful improvement of the signal to noise ratio by effective signal de-noising and extraction of the appropriate weak failure signs that may be obscured by other vibration sources and noise. The separation of bearing signals from other excitations sources is achieved at the signal processing stage. The purpose of the next stage, the feature extraction stage, is to compute reliable condition indicators. The article focuses on the feature extraction stage and the different methods for computing the condition indicators.

The condition indicators for bearing diagnostics are usually based on energies of peaks in envelope spectrum or statistical moments of the separated bearing signal. 

Several new methods for feature extraction of bearing damages are described. These methods are then compared and discussed. The advantages and disadvantages of each method for various scenarios are presented. The method comparison is performed using different cases of data from damaged bearings.

Ball bearing modeling for faults simulation


Today HUMS programs are based mainly on expensive, time consuming ground calibration tests. It is our goal to improve the modeling tools to reduce the time and budget needed to implement the HUMS approach.

A new 3D dynamic ball bearing model was developed. The aim of this generic model is to enable the dynamical response of a bearing with a wide spectrum of defects to be simulated to facilitate the development of condition indicators for bearing health status diagnostics and prognostics. Model validation includes a comparison between model outputs and known bearing response to local defects.

Experimental validation of the model to structural anomalies will be the focus of further research. A full scale rig for the validation of structural anomalies is under construction. The paper presents the description of the bearing model and the results of validation of the local faults. In addition, results for structural anomalies in the outer ring are presented. The sensitivity of the response to parameters such as load size, radial clearance and defect parameters is examined.

Condition Indicators for Gears


Diagnostics of faults in gears requires development of reliable condition indicators. A large number of condition indicators, which are based on statistical moments of the synchronous average and its derivatives (difference and residual signals) were previously suggested. This study evaluates the efficiency of different gear condition indicators that are based on statistical moments and compares them with two new types of condition indicators that are suggested. The two new types of condition indicators are based on the order spectrum and the spectral kurtosis of the synchronous average.

The study was conducted on the labeled data of PHM'09 challenge. This data included recordings of vibrations in helical and spur gearboxes with seeded faults.

Towards Model Based Prognostics - Characterization of Fault Size in Bearings


The characterization of a bearing fault and its size is a first important stage towards a reliable model based prognostics, leading to an assessment of the remaining useful life of the faulty bearing.

This paper describes the first stage in a research aimed to propose an approach for model-based prognostics.  

The approach utilizes various methods of vibration analysis and vibration based condition indicators which proved to be effective for detecting bearing degradation. While the problem of detecting a fault is more researched, the relationship between the health indicators (aggregated condition indicators) and the size of the damage, as well as the modeling of evolving damages are less researched and less understood. As a result, the ability to assess the severity of the damage or predict the remaining useful lifetime of a damaged bearing is limited. The current research is aimed to advance the knowledge in this difficult problem.

A 3D ball-bearing model that enables a simulation of its dynamics under the influence of a wide spectrum of faults was used. The goal was to study and predict the influence of the bearing fault size and the load on the generated vibration pattern. Based on the obtained insights, several health indicators were considered and compared.

The relationship between the fault size and the selected health indicators will be demonstrated by presenting results from experiments on a test fixture with seeded faults of different sizes and different loads, as well as the corresponding outcome of the dynamic model of the bearing.

Comparison of Methods for Separating Excitation Sources in Rotating Machinery


Vibro-acoustic signatures are widely used for diagnostics of rotating machinery. Vibration based automatic diagnostics systems need to achieve a good separation between signals generated by different sources. The separation task may be challenging, since the effects of the different vibration sources often overlap.

In particular, there is a need to separate between natural frequencies of the structure and excitations resulting from the rotating components (signal pre-whitening), and there is a need to separate between  signals generated by asynchronous components like bearings and signals generated by cyclo-stationary components like gears. Several methods were proposed to achieve the above separation tasks. The present study compares between some of these methods. For pre-whitening the study compares between liftering of the high quefrencies and adaptive clutter separation. The method of adaptive clutter separation is suggested in this paper for the first time.  For separating between the asynchronous and the cyclo-stationary signals the study compares between two methods: liftering in the quefrency domain and dephase. 

The methods are compared using both simulated signals and real data.

Detection of structural deformation in CH-53 swashplate bearing


The purpose of this paper is to present the research approach for the development of an algorithm for detection of a failure of the CH-53 swashplate bearing external spacer. The failure causes a lack of support of the swashplate bearing, thus creating a deformation of the outer ring.

This study integrates the results of a new 3D dynamic model, developed for assessment of the defect pattern, and results from experiments. The research approach is planned in hierarchical phases. The experimental phases include a small scale specimen, full scale test rig, helicopter blades test facility and finally a CH-53 helicopter. The unique approach gradually simulates the real work environment of the swashplate bearing.

The first two experimental phases and their results are presented. The first experimental phase is conducted on a small scale specimen and the second phase on a full scale test rig. Model results indicate that the lack of support has a defect pattern in both the radial and axial directions. These results are validated with the small scale specimen.

In the future phases, the algorithm will be validated with data from the helicopter blades test facility and CH-53 helicopter.

A method for anomaly detection for non-stationary vibration signatures


Vibration signatures contain information regarding the health status of the machine components. One approach to assess the health of the components is to search systematically for a list of specific failure patterns, based on the physical specifications of the known components (e.g. the physical specifications of the bearings, the gearwheels or the shafts). It is possible to do so, since the manifestation of the possible failures in the vibration signature is known a priory. The problem is that such a list is not comprehensive, and may not cover all possible failures. The manifestation of some failure modes in the vibration signature may be less investigated or even unknown. In addition, when more than one component is malfunctioning, unexpected patterns may be generated. Anomaly detection tackles the more general problem: How can one determine that the vibration signatures indicate abnormal functioning when the specifics of the abnormal functioning or its manifestation in the vibration signatures are not known a priori?  In essence, anomaly detection completes the diagnostics of the predefined failure modes. In many complex machines (e.g. turbofan engines), the task of anomaly detection is further complicated by the fact that changes in operating conditions influence the vibration sources and change the frequency and amplitude characteristics of the signals, making them non-stationary. Because of that, joint time-frequency representations of the signals are desired. This is different from other vibration based diagnostic techniques, which are designated for stationary signals, and often focus on either the time domain or the frequency domain.

For the purpose of this article, we will refer as TFR (time-frequency representation) to all 3D representations which employ on one axis either time, or cycles, or RPM, and on the other axis either frequency, or order. The proposed method suggests a solution for anomaly detection by analysis of various TFRs of the vibration signals (primarily the RPM-order domain).
In the first stage, TFRs of healthy machines are used to create a baseline. The TFRs can be obtained using various methods (Wigner-Ville, wavelets, STFT, etc). In the next stage, the distance TFR between the inspected recording and the baseline is computed. In the third stage, the distance TFR is analyzed and the exceptional regions in the TFR are found and characterized. A basic classification of the anomaly type is suggested. The different stages of analysis: creating baselines, computing the distance TFR, identifying the exception regions, are illustrated with actual data.


Subscribe to download publications



Copyright © R.K. Diagnostics 2009. All Rights Reserved. CFM Engine photo by Eric Drouin courtesy of Snecma