 Home |
|  Home |
3271 KB
AE SIGNALS FROM PROCESS MONITORING.Mr Phillip Cole (Physical Acoustics) Mr Timothy Bradshaw (Physical Acoustics) KEYWORDS Acoustic Emission, Process Monitoring. ABSTRACT: Acoustic Emission (AE) has been applied to the monitoring of process mechanisms over the last 50 years to provide information on the condition of the equipment involved or the process being conducted. This paper will consider the application of current AE technology to bearings and compressors. How and why AE is being applied to each application will be investigated including the proposed and achieved benefits provided by the results. INTRODUCTION The desire to acquire and record AE signals over the last 50 years has led to the development of sophisticated acquisition systems. These systems have been advanced by the rapid progression of computing technology enabling them to become smaller, faster and perform many more functions. From early systems acquiring the raw radio frequency (RF) signal through analogue processing into the digital age, the systems have evolved away from recording the RF signal towards acquiring smaller data files by pre-processing the signals as they arrive and extracting features by which these can be described. The benefit of this in terms of data file size, ease of data analysis, data filtering and process control has been widespread. There has always been a desire in some quarters to maintain the link to the raw signal in case any data is missed by feature extraction methods. This paper aims to overview how current AE technology is responding to these requirements as well as looking at some examples of simple process monitoring and what the different data acquisition techniques brings to the data analysis. SIGNAL PROCESSING To establish the differences between the two acquisition methods discussed in the introduction a Physical Acoustics PCI-2 AE data acquisition system controlled by Physical Acoustics AEwin acquisition software was used to monitor a simulated transient AE source. This source was generated by dropping an 8mm diameter steel ball bearing from a height of 100mm onto a concrete slab. A Physical Acoustics R6I sensor (40kHz to 369 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 100kHz frequency range) was placed at 1m distance from the point of impact with an integral pre-amp providing 40dB gain. The AE acquisition system was used to acquire a ten second sample of RF data at a rate of 5 million samples per-second (5MSPS). The results of this are displayed in Figure 1, and from the RF data each impact of the ball bearing can be clearly seen. The PCI-2 acquisition board was also used to acquire feature extracted data. As discussed in the introduction, in this case, when the RF signal passes a set threshold the transient is analysed and described by pre-defined features such as the amplitude, duration, rise time, absolute energy and counts of the signal. The AE acquisition system was used to acquire a ten second data file extracting features for transients over a 45dBAE threshold. The extracted amplitude results are displayed in Figure 2 and again each impact can be clearly identified. A clear benefit of the feature extracted data is that due to the dBAE amplitude scale it is possible to view on one scale very large hits as well as very small hits. Due to the linear nature of the RF scale very small hits are hard to see. A not so obvious benefit of the feature extraction is clear when you look at the file sizes generated during the 10 second of monitoring. This test has a very low hit rate, totalling 26hits. The RF data file is 99,091kB in size, while the hit data file is only 322kB in size. This means that the hit data file is 0.3% the size of the RF data file. This is a very important benefit as when monitoring a material or structure and nothing is happening, the RF data file will still be increasing in size but the feature extracted data file will only increase when emissions occur. For long tests this is a significant advantage. As discussed in the introduction, the RF data does hold all the information for the test and allows re-analysis of the data. Once features are extracted for a signal it is impossible go completely back to the original RF data. To date it has been hard to acquire both the RF and the feature extracted data simultaneously. Two acquisition systems have been required and different measurement chains used making direct comparison of the data against each other very difficult. However with the advances in AE acquisition by Physical Acoustics this has changed. Figure 3 displays the data displayed in Figures 1 and 2 directly against each other. It can be seen there is a direct correlation between the data files, as these were acquired simultaneously by the PCI-2 data acquisition system. The system is able to acquire simultaneous feature extracted data, time driven data, discreet waveform and parametric data onto one data file while also recording the RF signal data onto another data file. This test was repeated twice more using material samples that would produce real transient AE signals. A sample of composite material was monitored while a tensile load was applied. A Physical Acoustics Pico sensor (200 KHz to 750 KHz frequency bandwidth) was glued to the sample and an external pre-amplifier used to apply a 40dB signal gain (Figure 4). The direct comparison of this data is displayed in Figure 5. A sample of concrete material being exposed to a compressive point load was also tested. A Physical Acoustics R6I (40 KHz to 100 KHz frequency bandwidth) was glued to the sample and an integral pre-amplifier used to apply a 40dB signal gain, Figure 6. The direct 370 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 comparison of this data is displayed in Figure 7. These two materials were selected due to the common use of AE in the testing of these materials. The concrete test was the most emissive with 1025 hits being acquired. This increased the size of the data file to ~10% of that of the RF data file. As the settings were the same the RF data file remained the same size, 99,091kB while the feature extracted data file reached 10,358kB. Included in this feature extracted data file are also discreet waveforms for each hit. In Figure 8 the discreet waveform for one of the hits has been displayed while the RF data has been reduced to display the same hit. As can be seen, both are identical as you would expect. As tests on the concrete and composite materials are performed regularly the benefits of the feature extraction method are well understood. The RF files, as displayed so far, tell very little initially and take time to analyse. The feature extracted data can be graphed in many ways and used instantly to provide information on the type of damage occouring in the material and the severity of the damage mechanisms involved. All this is done real time and can be automated to provide feedback and alarm systems. One major advantage of the feature extraction system is the ability to do real time source location. As a demonstration, two Physical Acoustics R15I sensors (70 kHz-200 kHz) with internal 40dB gain were attached 1500mm apart to an aluminium box section 1800mm long. A wave velocity of 5000m/s was used to generate a linear location array locating events from their time of threshold crossing. A Hsu-Nielsen (HN) source (2H, 0.5mm diameter) was then used to simulate a source at one quarter, half and three quarters distance between the sensors on channels 1 and 2. Figure 9 displays the location results generated real time within the AEwin software. Figure 10 displays the RF data recorded on each channel as each HN source was made. Figure 11 displays the data from both sensors overlaid on each other. The RF data has been reduced to look at one HN source (one quarter distance) and it can be clearly seen, as would be expected, that the closest sensor, channel 1, receives the signal first while the sensor further away, channel 2, receives the signal at a later time. The data contained here in the RF signal concerns signal transfer times, modes and frequencies. The advantage of this type of data acquisition is that it allows fundamental analysis work to be conducted on the raw data. PROCESS MONITORING RESULTS AND DISCUSSION Having established the benefits of the two acquisition methods, the PCI-2 acquisition system was set up to monitor some process applications. The first of these was the monitoring of a bearing test rig, Figure 12. AE has been used widely on bearings due to its ability to give early warnings about the onset of damage and offers greater signal to noise performance over vibration methods. The rig consisted of two roller bearings 150mm apart on a balanced shaft driven by a belt from an electric motor. Physical Acoustics Pico sensors (200 KHz to 750 KHz frequency bandwidth) were glued to each bearing mounting and external pre-amps used to apply 20dB signal gain. Initially a HN source was used at bearing 1 to evaluate the signal transfer between the sensors. The data displayed in 371 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Figure 13 demonstrates that the signal is able to transfer between the sensors but has suffered large attenuation. The shaft was rotated at ~50RPM, the RF data acquired is displayed in Figure 14. By overlaying the signals from the sensors it is possible to see the dramatic difference between the two. Bearing 2 is significantly more active than bearing 1. This is the result expected as bearing 1 is a well oiled and maintained, while bearing 2 has been allowed to run dry and loose. There is still some emission from bearing 1 but by looking at the RF data closely (Figure 15) it is possible to see that it originates from bearing 2 and has been detected at bearing 1 as 'cross talk'. This can be assumed as the signal transfer time is close to that seen from the HN source ~100usec. The simultaneous feature extracted data is displayed in Figure 16. Again this is a very clear result with bearing 1 being significantly quieter than bearing 2. To allow very simple data acquisition the Root mean square (RMS) of the RF signal was also calculated every 10milli seconds, the result is displayed in Figure 17. This also provides the same result but without requiring complicated acquisition systems. This demonstration shows that the RF and feature extracted data provides the ability to do some analysis on the signals, but if a simple good / bad diagnosis is required the RMS calculation will be more than adequate. Small on line monitoring systems are available with alarm functions that will identify a bad bearing without the need to store any data at all. The second process application to be investigated is the monitoring of a compressor valve shown in Figure 18. Again AE has been applied in this area due to the signal to noise ratio benefits over vibration monitoring and ability to detect and diagnose problems within the compressor. A Physical Acoustics Nano 30 sensor (125 KHz to 750 KHz frequency bandwidth) was attached to the valve mounting and an external pre-amp used to apply 20dB signal gain. The compressor is a single cylinder with only one inlet and one outlet valve. The cylinder is driven at ~50Hz by an electric motor. A ten second data RF data file was collected as displayed in Figure 19. By looking at this data 48 cycles of the compressor can be seen matching the ~50Hz performance of the machine. In Figure 20 the RF signal has been reduced to look at one cycle of the compressor. It was not possible to mount a top dead centre (TDC) marker on this particular compressor so in depth diagnosis on this data is not possible. If a TDC marker had been available it would have been able to correlate the repetitive emissions to particular events in the mechanical workings of the compressor (inlet valve open, inlet valve close, outlet valve open, outlet valve close). To simulate a problem with the compressor the air inlet valve was blocked open by a small obstruction. This did not prevent the flow of air into the compressor but prevented the valve closing fully, simulating a damaged valve. The RF data from one cycle of the compressor under these circumstances is displayed in Figure 21. As can be seen the signature of the compressor has changed and this result is mirrored in the RMS and 372 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 average signal level (ASL) results displayed in Figures 22 and 23. With the valve blocked the signal levels have actually dropped probably because the inlet valve can not slap open and closed as normal and there is not enough pressure generated to slap the outlet valve open and closed as is normal. This inhibits the performance of the compressor while reducing the amplitude of the AE signals generated. CONCLUSIONS This paper has looked at the advantages and disadvantages of both RF and feature extraction methods of AE data acquisition. It has been established that although the RF data produces very large data files that are cumbersome to analyse they do contain fundamental information that is essential when establishing the behaviours of AE in applications that may not be fully understood. The feature extraction data acquisition allows intelligent data acquisition, collecting data only when required. With the combination of the many features that can be extracted, complex data analysis can be conducted real time identifying under what conditions different AE source mechanisms are active and providing the ability to locate these sources on a material. The data can be used to feedback into control systems allowing alarms to be set making the AE system an intelligent system controller. The ability to provide a discreet picture of the RF signal for each hit also provides some flexibility in post-processing allowing more accurate locations to be achieved and some fundamental data obtained while still keeping file sizes to a minimum. As demonstrated in some applications the complexity of the RF or even the feature extracted data is not required. Time driven data providing simple RMS or ASL signal levels can be enough to diagnose a potential problem. The Physical Acoustics PCI-2 acquisition system (Figure 24) makes use of the most modern technology to enable it to be the first AE research tool to combine all these techniques into one complete system. The flexibility this provides to the user is unparalleled, allowing complete control over the form and volume of AE data collected. 373 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Initial ball bearing impact Ball bearing rebounds off concrete surface Figure 1: RF signal acquired for 8mm ball bearing dropped onto concrete surface from 100mm height. Initial ball bearing impact Ball bearing rebounds off concrete surface Figure 2: Extracted amplitudes for 8mm ball bearing dropped onto concrete surface from 100mm height. 374 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Direct correlation between RF and extracted data Figure 3: Direct comparison of RF signal and extracted amplitude acquired for 8mm ball bearing dropped onto concrete surface from 100mm height. 375 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 AE Sensor Figure 4: Composite tensile sample with sensor attached. RF data Direct correlation between RF and extracted data Extracted data Figure 5: Direct comparison of RF signal and extracted amplitude from composite sample under tension. 376 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 AE Sensor Figure 6: Concrete compression sample with sensor attached. RF data Direct correlation between RF and extracted data Extracted data Figure 7: Direct comparison of RF signal and extracted amplitude from concrete sample under compressive point load. 377 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Extracted data Discreet waveform RF data Figure 8: Comparison of discreet waveform acquisition and acquired RF signal. 378 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Linear position Sensor Sensor Linear position Figure 9: Location of HN source along sample. CH 1 Time CH 2 Time Figure 10: Comparison of RF signals from both AE sensors for linear location. 379 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 RF data HN signal arrival at CH 1 HN signal arrival at CH 2 Figure 11: Comparison of RF signals from both AE sensors from one HN source 1/4 of the way along the sample (closest to channel 1). 380 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 AE Sensor Bearing 1 Bearing 2 Figure 12: Bearing test rig with AE sensors mounted. RF data HN signal arrival at bearing 1 HN signal arrival at bearing 2 Figure 13: Example of signal transmission from one bearing to another with RF acquisition. 381 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Signals from bearing 2 Signals from bearing 1 RF data Figure 14: 10 seconds of RF data comparing AE from two bearings. Signal arrival at bearing 2 Signal arrival at bearing 1 RF data Figure 15: RF data comparing single AE event at two bearings. 382 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Bearing 1 Extracted data Extracted data Bearing 1 Extracted data Bearing 2 Extracted data Bearing 2 Figure 16: Extracted Amplitude results for each bearing. Figure 17: Time driven RMS results for each bearing. 383 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 AE Sensor Figure 18: Compressor with AE sensor attached 48 cycles (~50Hz compressor) Figure 19: 1 second of RF data from compressor. 384 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 RF data Figure 20: RF signal from one rotation of compressor. RF data Figure 21: RF signal from one rotation of compressor with inlet valve leak. 385 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 RMS ASL Figure 22: ASL and RMS data from compressor. RMS ASL Figure 23: ASL and RMS data from compressor with inlet valve leak. 386 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Figure 24: A Physical Acoustics PCI-2 data acquisition board. 387 DGZfP-Proceedings BB 90-CD Lecture 36 EWGAE 2004 Mr Timothy Bradshaw: Address: Physical Acoustics Ltd Norman Way Over Cambridge CB4 5QE UK Tel: +44(0)19 5423 1612 Fax: +44(0)19 5423 1102 E-mail: tpb@pacuk.co.uk Mr Phillip Cole: Address: Physical Acoustics Ltd Norman Way Over Cambridge CB4 5QE UK Tel: +44(0)19 5423 1612 Fax: +44(0)19 5423 1102 E-mail: ptc@pacuk.co.uk 388 |
| © NDT.net |