Automatic Bearing Fault Detection

GWOC uses the WABI-WABI to perform automated tests and measurements on rolls that aim at finding bearing defects. The analysis of the measurements is still done by GWOC personnel. This analysis is time consuming, and GWOC is not able to detect any other defects than outer ring defects. The main question in this report was whether it would be possible to automate the analysis of measurements (by a computer algorithm), and under what conditions this algorithm would perform better than the current GWOC method. An Automatic Bearing Fault Detection Algorithm, based on the HFRES technique has been developed. The performance was evaluated using 100 measurements of bearings. Evaluation on these measurements indicate that it will perform slightly better than the current GWOC method: in 43% of the cases it was conservative (against 46% with the current GWOC method). Not unimportantly, the work as described in this report has made it more clear how to improve the detection performance.

Introduction

In the realm of machinery reliability and performance, the timely detection of bearing faults stands as a pivotal factor in averting catastrophic failures and minimizing operational downtime. However, the effectiveness of fault detection algorithms is often contingent upon the intricacies of the testing environment. This paper delves into the development and evaluation of an Automatic Bearing Fault Detection Algorithm, specifically tailored to grapple with the challenges posed by the GWOC test setup.

The GWOC test configuration introduces unique complexities, notably a broad frequency range and variations in vibration patterns across distinct bearings. Addressing these challenges becomes imperative for accurate and reliable fault detection. The overarching objective of our study is to introduce an algorithm capable of not only replicating but also augmenting the bearing fault detection performance observed in the GWOC test conditions.

Therefore our research question is: Can the bearing fault detection be automated, such that the above requirements are satisfied, and if so, can it improve over the current GWOC method?

To answer this question, we have the following subquestions:

  • What are the necessary conditions for a successful implementation of the HFRES technique?
  • What is a suitable Automatic Bearing Fault Detection Algorithm, and are there possibilities to improve it (better performance, and ability to detect inner ring faults and ball bearing faults)?

The HFRES detection method

The High Frequency Resonance Envelope Spectrum (HFRES) technique stands out as a robust approach for monitoring vibration measurements, particularly in the context of detecting bearing defects. Its effectiveness lies in its ability to distinguish vibrations originating from a faulty bearing amidst the broader spectrum of mechanical elements. This success has established HFRES as a reliable method in various industries.

Understanding Bearing Defects

Before delving into the HFRES technique, it’s crucial to comprehend the common types of defects found in bearings. Defects can arise from fatigue, wear, poor installation, improper lubrication, and various other factors affecting components like the outer race, inner race, cage, and rolling element, see figure 1. Each of these defects manifests differently in vibration measurements, influencing the overall mechanical system.

Figure 1: Bearing components.

The HFRES approach relies on the concept of Defect Frequencies, derived from parameters like the Pitch Diameter and Rotating Frequency. These frequencies help identify defects in bearing components by analyzing the Envelope Spectrum.

Defect Frequencies

When a bearing rotates, it generates periodic frequencies known as Fundamental Defect Frequencies. The associated Defect Multipliers, such as Ball Pass Frequency of the Outer race (BPFO) and Inner race (BPFI), provide critical information for defect identification. The Envelope Spectrum, derived through AM demodulation, becomes instrumental in detecting defects by highlighting resonances within the frequency range of 10 Hz to 50 KHz.

However, practical considerations, such as mechanical misalignment or poor speed control, may lead to discrepancies between calculated and observed frequencies. Mechanical Looseness and its harmonics, often identified as harmonics of 1x the Rotating Frequency, further add complexity to the Envelope Spectrum.

HFRES Technique Steps

The HFRES technique involves two main blocks: AM Demodulation and Signal Analysis. The former processes vibration measurements to create the Envelope Spectrum, while the latter analyzes this spectrum for occurrences of Fundamental Defect Frequencies.

AM Demodulation encompasses band-pass filtering, envelope detection, and Fast Fourier Transform (FFT). Following this, Signal Analysis utilizes various techniques like narrow-band filtering and crest factor analysis to identify defects.

Implementation Conditions

Successful HFRES implementation requires specific conditions, including the absence of Mechanical Looseness, constant Rotating Frequency, known Resonance Frequencies, and well-mounted accelerometers. A pre-analysis of GWOC measurements confirms the presence of Mechanical Looseness and uncertainty in Rotating Frequency, highlighting the challenges in implementing the HFRES technique.

Pre-analysis

A set of 100 vibration measurements, specifically provided by GWOC to DotX, forms the basis for designing the Automatic Bearing Fault Detection Algorithm. Each of these measurements corresponds to bearings that underwent Visual Inspection, providing crucial feedback on the actual status of the bearings—whether they are faulty or in good condition. This integration of real-world feedback adds a layer of authenticity to the analysis, allowing for a more robust evaluation of the HFRES technique’s performance on bearings with known conditions.

The Envelope spectrum of two bearings were computed in order to determine whethere detect frequencies could be spotted in the low frequency spectrum. First, a bearing that is known to be good is analysed, see figure 2. Second, a bearing that is known as bad is analysed, see figure 3. We can see that the difference between a good and a faulty bearing is more obvious in the envelope spectrum.

Figure 2: Bearing with no defect.

Figure 3: Bearing with defect.

Conclusion of HFRES Detection Method

The analysis of GWOC measurements using the HFRES technique underscores the potential of this method in detecting bearing defects. The Envelope Spectrum proves to be a valuable tool, surpassing the limitations of the Original Spectrum in distinguishing between good and faulty bearings. However, challenges such as Mechanical Looseness and uncertainty in Rotating Frequency necessitate careful consideration for a reliable implementation.

The Automatic Bearing Fault Detection Algorithm

In this chapter, we delve into the intricate design of the Automatic Bearing Fault Detection Algorithm utilizing the High-Frequency Resonance Envelope Spectrum (HFRES) technique. The prerequisites for the successful implementation of the HFRES technique are established. The analysis presented in Chapter 2 highlights that the GWOC test setup fails to comply with these conditions. Nevertheless, the development of an algorithm capable of either replicating or surpassing the GWOC Bearing Fault Detection performance in the existing test setup conditions is deemed feasible. To achieve this, several steps are proposed in the algorithm design to accommodate the limitations of the test setup, see a overview of the algorithm in figure 4.

Figure 4: Automatic Bearing Fault Detection.

AM Demodulation

The AM Demodulation block contains a bandpass filter from 100 Hz to 8000 Hz, outputs the absolute value of the band pass filtered signal, a low pass filter at 5.5 timer the largest defect frequency to include at least 5 harmonics and a FFT of the low pass filtered signal.

Envelope Spectrum Processing

Prior to Signal Analysis, enhancements in the Envelope Spectrum are proposed through preprocessing. A band-pass filter is applied directly to the spectrum to eliminate bias and noise. Standard filters, though effective, tend to distort phase, which is unacceptable for this application. Therefore, a non-causal filter is employed to maintain the integrity of the Envelope Spectrum, ensuring accurate defect peak detection. See in figure 5 the original Envelope Spectrum and the processed Envelope Spectrum.

Figure 5: Original and processed Envelope Spectrum.

Peak Detection and Defect Search

The algorithm incorporates a two-step process for peak detection and defect search. The first step identifies high-amplitude peaks in the processed Envelope Spectrum. Relevant peaks are selected using the ‘peakdet’ algorithm, distinguishing between high-amplitude peaks and low-amplitude noise. An initial selection of bearings as good, faulty, or undecided is made based on the amplitude differences between positive and negative peaks.

The second step involves the Defect Search algorithm, which utilizes defect multipliers and Rotating Frequency to identify potential defects in the bearing. The algorithm outputs the peaks found for each Defect Frequency harmonic, facilitating further analysis.

Decision Making

The Decision Making block refines the results from the Defect Search algorithm. It deletes occurrences of Rotating Frequency peaks that could be attributed to Mechanical Looseness. If any of the vectors related to specific defect frequencies (BSF, BPFO, or BPFI) contains more than one element, the bearing is diagnosed as faulty; otherwise, it is deemed good. This step ensures accurate diagnosis, considering the harmonics inherent in defects.

Tuning Parameters

Tuning parameters play a crucial role in optimizing the algorithm’s performance. The ’threshold’ factor in the Peak Detection step can be adjusted to include more or fewer peaks in the analysis, impacting Conservativeness. If Mechanical Looseness is expected, a decrease in the threshold may be beneficial.

The maxdev value in the Defect Search step accounts for the uncertainty in Rotating Frequency. Adjusting maxdev based on the expected uncertainty helps fine-tune Conservativeness. During the algorithm’s design, these tuning parameters were meticulously set to achieve optimal performance.

The overall solution block diagram is illustrated in Figure 3.1, showcasing the intricate interplay of components in the Automatic Bearing Fault Detection Algorithm. This algorithm, designed to navigate the complexities of the GWOC test setup, stands poised to deliver robust and accurate results in the realm of bearing fault detection.

Results

The Automatic Bearing Fault Detection Algorithm’s performance is assessed against the identical measurement set used for GWOC Bearing Fault Detection. The obtained performance indexes, with the specified tuning parameter values, are displayed in Table 4.1.

GWOCAutomatic Bearing Fault Detection Algorithm
Conservativeness46.15%43.26%
Failure0.96%0%
Table 4.1: Automatic Bearing Fault Detection Algorithm Performance.

Conclusion

In summary, the Automatic Bearing Fault Detection Algorithm, designed based on the HFRES technique, demonstrates competitive performance in comparison to the GWOC standard. The careful consideration of tuning parameters allows for customization, balancing conservativeness and failure rates, but highlights the algorithm’s sensitivity to uncertainties in Rotating Frequency and Mechanical Looseness.

Plaats een reactie

nl_NLDutch