NDTnet - November 1997, Vol.2 No.11
A DISTRIBUTED ARCHITECTURE FOR NON-DESTRUCTIVE INSPECTION OF LARGE AIRCRAFT PARTS
A. Ibáñez, et al *
Abstract
Advanced composite structural parts are continually expanding in aircraft industry. Automatic NDI of such parts must fulfill very strict specifications, especially related to quality of images and inspection speed. Multi-channel operation and digital signal processing provide a good solution to overcome such inspection demands. In this paper a distributed architecture for automatic NDI by ultrasound is presented. A central computer can drive several multi-channel remote systems (R-SENDAS1), which are located near the transducers. Each R-SENDAS makes, at a global rate of 10 Msamples/s, digital processing of ultrasonic (UT) signals for SNR enhancement, resolution improvement and data reduction. Processing algorithms can be on line configured for through-transmission or pulse-echo techniques. Data reduced containing all the relevant information for the evaluation task, are transmitted to the central computer through a 5Mb/s, point to point, serial communication line. The central computer, free from UT attention, is dedicated to other tasks, such as movement control of the inspecting robot, UT image display, etc.
1 "SENDAS" stands from the spanish "Sistema de END de Arquitectura Segmentada", and "R" means "Remote".
TABLE OF CONTENTS
Introduction
Automatic UT-NDI of aircraft parts is frequently carried out by means of a scanning robot, which is remotely located from the evaluation site
(figure 1).
A central computer controls the robot movements and drives ultrasonic acquisitions. NDI of big structural parts requires large size scanning devices, performing complex trajectories involving many-controlled axis. To increase productivity, such equipment needs multi-channel configurations that provide several lines per scan. Usually signals are sent from the acquisition point to the control site in analog form, perhaps after passing through head amplifiers, requiring long shielded cables to reach the computer, where they are digitized and processed. Apart from the cable count and installation complexity, this method has
the shortcoming of noise sensitivity, which can degrade the dynamic range and signal integrity. This is specially relevant in large scanning robots where, both, long wiring and high power motors are involved producing strong spikes of electromagnetic (EMI) noise which cannot be easily removed by analog techniques when weak signals are involved. To overcome this drawback, it would be possible to digitize the signals so close to the transducers as possible and to use high-speed digital data links to reach the evaluation computer. However this demands a quite large bandwidth to keep the high scanning speed required in production processes, where data rates of several Mbytes/s are very common.
Distributed architecture
In our approach, a multi-channel remote system R-SENDAS locally performs both signal acquisition and digital processing. In the final step, reduced data are obtained, which contain all the relevant information for the evaluation task (
figure 2). R-SENDAS follows the approach of SENDAS architecture [1]. It performs complex algorithms for noise suppression, resolution improvement, and smart data reduction, by means of high speed dedicated processors, at a sustained global 10 Msamples/s processing rate. Data are transmitted to the central computer by a full duplex, 5 Mb/s, point to point communication. The central computer can drive several R-SENDAS at distances above 200 m. in a two-level tree topology ( figure 3), in which the remote stations perform ultrasonic acquisitions and signal processing, while the central computer carries out transducers positioning and acquisition synchronization.
This approach has several advantages, especially for aeronautic NDI applications where a great number of transducers, which are located far from the central site, can be involved. First, the simplicity of wiring reduces the installation costs (a single coaxial cable per remote station). Second, it provides quite high noise immunity by reducing the length of signal paths. And third, a highly efficient distributed configuration is obtained, in which complex digital algorithms are locally performed in real time without consuming processor time from the central computer. It is worth to highlight the hierarchical structure of the overall system with different levels of parallel processing, which allows high data processing rates.
Remote SENDAS structure
The main goal on the design of R-SENDAS has been to obtain high quality acoustic images at low cost. Figure 4 shows a block diagram of R-SENDAS architecture that is based on the SENDAS concept [1,2]. This is a mixed signal architecture, which supports both analog conditioning and digital processing of signals up to eight multiplexed channels. Digital modules operate over the data at a sustained 10 Msamples/s processing rate, performing hardware implemented DSP algorithms. Although digital processing is made by hardware, flexibility is kept at three levels:
- The modules installed in the pipeline, each one executing a specific function, define algorithms.
- Functionality of most modules can be in-process modified, by reconfiguring the hardware. To this purpose, programmable FPGA devices allow changing circuits by loading different configuration files.
- At the lower level, modules have several programmable parameters, which can be tuned to the application requirements.
A detailed information of SENDAS architecture and its processing modules can be found elsewhere in this journal [2].
A local processor carries out the remote management by communicating with the central computer, and programming and controlling the processing modules. For setup purposes, a memory is connected to the signal bus allowing the capture and transmission of the data coming from any processing module ( fig 4). This way, the system parameters such as time windowing, DAC curves, filters tuning, data reduction algorithms, etc. can be configured and optimized during the setup stage.
Remote SENDAS Operation
The local processor of R-SENDAS disposes of internal memory to save configuration parameters of every channel independently, therefore, the configuration can be established during the setup stage and the central computer does not requires any care of this task during acquisitions. When an UT acquisition is required, R-SENDAS carries out all the needed tasks:
(a) Programming modules and timers with the parameters of the selected channel, and (b) firing the UT pulse, processing the signals and transmitting the final results, which can be complete traces (for setup or calibrating purposes) or reduced data (for automatic NDE). In an automatic inspection, the central computer controls the synchronism of acquisitions by requiring new UT data when the scanning robot surpasses every resolution cell.
An analysis of the basic timing of SENDAS architecture can be found in [2]. R-SENDAS operates sequentially over each channel, and the time for each channel includes the following:
- Programming time Tp devoted to configure the modules and timers with the parameters of the selected channel; it is a fixed time under 100ms.
- Inhibition time Td, which is the interval from the transducer firing to the start of data acquisition. During this interval, no relevant information is present at the system input. When Td >Tp, the programming time is transparent since this function can be carried out during the inhibition interval.
- Digital Processing time Tdp , which is 0.1ms*N, being N the number of samples of the UT trace. If the sampling frequency is lower than the clock of the signal bus (10 MHz), then Tdp equals the acquisition time Tz, in other case, the loss of time due to digital processing is:
Tdp-Tz = N/10-N/Fs (in ms)
being Fs the sampling frequency in MHz. And,
- Transmission time Ttdevoted to communicate data to the central computer. Serial transmission of a byte takes 2 ms. However, the time spent for data management is clearly longer: 6ms/data when complete traces are transmitted (120.000 data/s), or 13 ms/data when reduced data are sent.
R-SENDAS can be configured for through-transmission or pulse-echo operation:
(a) Through-transmission
- In through-transmission only a C-scan image is needed with a colored or gray scale representation of the amplitude of the UT signal passing through the tested part. A typical configuration for such inspections is shown in figure 5. The UT signal coming from the selected transducer (Ti) is analog conditioned by a logarithmic amplifier, which covers a dynamic range of 100 dB. Once the inhibition time has elapsed, digitization and digital processing of the UT signal starts at the pipeline.
- Certain high-attenuating materials (this is the case of thick honeycomb sandwich parts) demand very high dynamic range near 100 dB. In such cases, EMI noise from the driving motors must be digitally removed. Spatial filtering applied over consecutive traces can easily reduce such impulsive noise, especially when non-linear algorithms are performed [3]. Figure 6) shows comparative examples of EMI noise reduction in a through-transmission experiment by using both linear and non-linear algorithms. The greater efficiency of the second case is evidenced.
- After EMI noise suppression, data are fed to the Data Reduction module. For C-scan images, only the peak value of the UT signal during a windowed interval is necessary. Therefore, after configuring the module as Single Peak Detection, this computes the maximum amplitude of the signal, which is transmitted to the central computer.
- Figure 7 shows the C-scan image from a honeycomb nomex core part whose thickness varies from 0 to 150 mm, which has been acquired by R-SENDAS. Emitting pulses at a central frequency of 1 MHz, the part produces a drop of about 75 dB. In addition, 20 dB are still required to evaluate defects in the maximum attenuation area. This leads to a dynamic range of about 95 dB that rather exceeds the range of most commercially available systems.
- In figure 8) a timing diagram is shown for a trace with 2000 samples (200 ms acquisition time, sampling at 10 Msamples/s), where an inhibition time of 100 ms has been considered. The digital processing interval is 200 ms and the transmission time for a single data is 15 ms. A 4-channel system will take 1250 ms, 5% of the interval being spent in data transmission. For 1mm resolution, the scanning speed of the robot can reach 800 mm/s, which yields inspection rates above 10 m²/hour.
Pulse-echo operation
- The high processing ability of SENDAS can be advantageously used for pulse-echo applications, in which complex algorithms for SNR improvement and data reduction can be simultaneously performed at high speed. Two kinds of arrangements can be used in pulse-echo:
- The simplest one is based on analog envelope detection and further digital processing over the video trace figure 9. After digital conversion, a module for EMI noise suppression by non-linear filtering can be included. The drastic effect of this filter over the UT signal can be observed in figure 10, where noise spikes are above the signal level. Then, a multi-peak detector, which captures the amplitude (1 byte) and position (2 bytes) of successive echoes surpassing a programmable threshold, gives a suitable data reduction. An objection to this method is that, by extracting the video trace, it does not use all the radio frequency (RF) information of signals, in order to improve SNR or depth resolution before data reduction.
- Figure 11 shows the arrangement in the case of processing RF signals. After digital conversion, spatial filtering with a linear algorithm may be advisable in order to improve SNR. Once the signal is filtered, a deconvolution process [4] helps to increase the depth resolution. As the output of the deconvolution module is a RF signal, digital envelope detection [5] must be computed by means of a new module, before the multi-peak detection is performed. As this configuration employs more than four processing modules, a complementary base card has to be coupled to the main one. Several examples operating with RF signals are shown in reference [2].
- A timing diagram for UT traces of 2000 samples, Td being 100 ms, is shown in figure 12. The digital processing interval Tdp remains in 200 ms. It has been observed that an averaged value of the number of detected peaks in most pulse-echo applications is about three, which are codified in 3*3=9 bytes. The transmission time Tt is about 180 ms per UT pulse. A multi-channel system with 4 channels will take 1920 ms, 35% of the interval being spent in communications. For 1mm resolution, and a scanning speed of the robot of 250 mm/s, the inspection rate is 4 m²/hour.
Conclusions
Remote digital processing of ultrasonic signals offers several advantages over other methods in different aspects. Obviously cabling and installation costs are reduced. More important is the noise immunity given by this approach, that allows operating with a dynamic range near 100dB even in noisy ambient conditions. Furthermore, a distributed multi-remote architecture can be configured to increase inspection productivity. Finally by incorporating the SENDAS digital pipeline of processing modules to this concept, a high performance ultrasonic inspection system for automated NDE is obtained. Such advantages lead to an easy incorporation of multi-channel pulse-echo techniques to automatic NDI of aeronautic composite parts.
References
- C.Fritsch et al., A pipelined architecture for high speed automated NDE, Proceeding 1995 IEEE Ultrasonic Symposium, pp. 833-836, 1995.
- C.Fritsch et al., Digital Signal Processing of ultrasonic NDE signals in hard real-time environments: the SENDAS approach., in this UT on-line Journal.
- I. Pitas, A.N. Venetsanopoulos, Nonlinear Digital Filters, Kluwer Academic Pub., Norwell, Mass. 02061, USA, 1995.
- J.J. Anaya, L.G. Ullate, C. Fritsch: A Method for Real-Time Deconvolution. IEEE Trans. Instrum. and Meas., IM-41, 3, pp. 413-419, june 1992.
- C. Fritsch, A. Ibañez, M. Parrilla, A digital envelope detection filter for Real-Time operation, IEEE Trans. on Instrumentation and Measurement, (to be accepted).
ACKNOWLEDGMENTS
This work has been supported by: CICYT grant TAP 94-0812 of the Spanish Ministry for Science and Education, CAM grant no. I+D 52/94 of the Community of Madrid, PACE project PC144 of the European Community - Spanish Ministry for Industry and Energy, and the company TECAL S.A., currently manufacturing the SENDAS architecture.
AUTHOR
|Top to this page|
For more information see: NDT in Aerospace - UTonline 11/97
© Copyright 1. Nov 1997 Rolf Diederichs,
rd@ndt.net
/DB:Article /AU:Ibáñez_A /AU:et_al /CN:ES /CT:UT /CT:instrument /CT:composite /CT:aerospace /ED:1997-11
figure 2). R-SENDAS follows the approach of SENDAS architecture [1]. It performs complex algorithms for noise suppression, resolution improvement, and smart data reduction, by means of high speed dedicated processors, at a sustained global 10 Msamples/s processing rate. Data are transmitted to the central computer by a full duplex, 5 Mb/s, point to point communication. The central computer can drive several R-SENDAS at distances above 200 m. in a two-level tree topology (
figure 3), in which the remote stations perform ultrasonic acquisitions and signal processing, while the central computer carries out transducers positioning and acquisition synchronization.
Figure 4 shows a block diagram of R-SENDAS architecture that is based on the SENDAS concept [1,2]. This is a mixed signal architecture, which supports both analog conditioning and digital processing of signals up to eight multiplexed channels. Digital modules operate over the data at a sustained 10 Msamples/s processing rate, performing hardware implemented DSP algorithms. Although digital processing is made by hardware, flexibility is kept at three levels:
figure 5. The UT signal coming from the selected transducer (Ti) is analog conditioned by a logarithmic amplifier, which covers a dynamic range of 100 dB. Once the inhibition time has elapsed, digitization and digital processing of the UT signal starts at the pipeline.
Figure 6) shows comparative examples of EMI noise reduction in a through-transmission experiment by using both linear and non-linear algorithms. The greater efficiency of the second case is evidenced.
Figure 7 shows the C-scan image from a honeycomb nomex core part whose thickness varies from 0 to 150 mm, which has been acquired by R-SENDAS. Emitting pulses at a central frequency of 1 MHz, the part produces a drop of about 75 dB. In addition, 20 dB are still required to evaluate defects in the maximum attenuation area. This leads to a dynamic range of about 95 dB that rather exceeds the range of most commercially available systems.
figure 8) a timing diagram is shown for a trace with 2000 samples (200 ms acquisition time, sampling at 10 Msamples/s), where an inhibition time of 100 ms has been considered. The digital processing interval is 200 ms and the transmission time for a single data is 15 ms. A 4-channel system will take 1250 ms, 5% of the interval being spent in data transmission. For 1mm resolution, the scanning speed of the robot can reach 800 mm/s, which yields inspection rates above 10 m²/hour.
figure 9. After digital conversion, a module for EMI noise suppression by non-linear filtering can be included. The drastic effect of this filter over the UT signal can be observed in 
figure 10, where noise spikes are above the signal level. Then, a multi-peak detector, which captures the amplitude (1 byte) and position (2 bytes) of successive echoes surpassing a programmable threshold, gives a suitable data reduction. An objection to this method is that, by extracting the video trace, it does not use all the radio frequency (RF) information of signals, in order to improve SNR or depth resolution before data reduction.
Figure 11 shows the arrangement in the case of processing RF signals. After digital conversion, spatial filtering with a linear algorithm may be advisable in order to improve SNR. Once the signal is filtered, a deconvolution process [4] helps to increase the depth resolution. As the output of the deconvolution module is a RF signal, digital envelope detection [5] must be computed by means of a new module, before the multi-peak detection is performed. As this configuration employs more than four processing modules, a complementary base card has to be coupled to the main one. Several examples operating with RF signals are shown in reference [2].
figure 12. The digital processing interval Tdp remains in 200 ms. It has been observed that an averaged value of the number of detected peaks in most pulse-echo applications is about three, which are codified in 3*3=9 bytes. The transmission time Tt is about 180 ms per UT pulse. A multi-channel system with 4 channels will take 1920 ms, 35% of the interval being spent in communications. For 1mm resolution, and a scanning speed of the robot of 250 mm/s, the inspection rate is 4 m²/hour.