02-10-2012, 05:38 PM
Signal Processing for Ultrasonic Testing of Material with Coarse Structure
Signal Processing.docx (Size: 178.88 KB / Downloads: 25)
In pulse-echo inspection the backscattering from the grain boundaries appears in the received ultrasonic images as clutter, often referred to as grain noise. Several processing algorithms for signal enhancement have been proposed, of which the Split Spectrum Processing (SSP) probably is the most renowned. The SSP technique applies a filter bank to some frequency band that has to be precisely known in advance, to obtain a set of signals with decorrelated noise components.
Two novel algorithms are presented and verified in the paper. The first algorithm is based on the concept of SSP but it does not require a priori knowledge of the frequency range for locating the SSP´s filter bank. The algorithm referred to as Consecutive Polarity Coincidence (CPC), makes explicit use of the pulse characteristics of the target echo in order to implement local bandwidth estimation. A threshold can be set, equal to the frequency range utilised for processing that will generate a gating signal identical to the one obtained when using conventional Polarity Thresholding.
The second algorithm is based on a completely different approach that assumes using a transducer with as wide as possible frequency range and then estimating noise and target parameters to adapt the effective frequency range to the properties of inspected material. The resulting algorithm is a modified version of noncoherent detection (NCD) which is known from communications. To adapt the algorithm for the use in ultrasonic NDT a two-parameter model of ultrasonic wavelets is used.
The performance of the algorithms, especially their temporal resolution, is compared and illustrated using ultrasonic images acquired from a specimen made of cast stainless steel with material structure is similar to that found in nuclear power plants.
Introduction
Ultrasonic non-destructive inspection of coarse grain materials, such as stainless steel or copper, can often be complicated by the presence of backscattering from the material microstructure. It is well known that this type of noise can, in many cases, be substantially reduced by means of time-frequency signal processing algorithms. The most commonly used algorithms in this group are Split Spectrum Processing (SSP) techniques, [1]. These techniques share a common structure, including a filter bank for signal expansion and a non-linear statistical processor for target echo extraction. The filter bank splits the original broadband ultrasonic signal into a number of small-band signals. The SSP principle is based on a hypothesis that for each time instant small-band components of the grain noise are more or less decorrelated, while a large degree of correlation is observed for target echoes. Then target echo extraction may be implemented by applying a suitable measure of correlation to the generated set of small-band signals. The SSP approach uses for signal expansion a filter bank consisting of Gaussian band-pass filters overlapping in frequency domain. Simple target extractors such as the Polarity Thresholding and the Amplitude Minimization algorithms have been suggested and proven successful provided that the processing parameters had been correctly tuned [2].
However, SSP is sensitive to two key parameters, centre frequency and bandwidth of the processing range [3]. Even a slight deviation in the frequency range may result in target echo cancellation, since a single split signal containing no target information is sufficient for violating the algorithm operation principle [2]. Consequently, good a priori knowledge of the target echo spectrum is required for achieving adequate results.
One solution to the problem is using robust techniques that are suitable for situations where the exact frequency location of the target is not known is advance, but it can be presumed that the target spectrum occupies a considerable frequency range of the received spectrum. Generally, for adequate target extraction, both the frequency selectivity of the Split Spectrum approach and the ability to dynamically adapt the frequency range used for target extraction to the local properties of the received signal are required. An extraction algorithm satisfying these requirements, referred to as Consecutive Polarity Coincidence (CPC), is presented in the next section
Consecutive polarity coincidence
The objective of adaptive frequency range, mentioned above, can be met by using a slightly different interpretation of the Polarity Thresholding (PT). PT is based on the assumption that the ultrasonic target echo would have features of a mathematical pulse, which means that for a specified frequency band the echo would be composed of in-phase Fourier components. Time-frequency representation of the received ultrasonic signal obtained from the split spectrum filter bank is subjected to a simple polarity test. The conventional PT detection algorithm indicates a target echo at those instants of time when all the split signals share a common polarity. Obviously, the underlying assumption can be very easily violated, by adding a single split signal with the opposite polarity.
Noncoherent Detector (NCD)
The noncoherent detector, which is well known from telecommunication, is designed for detection of bandpass signals, s(t)=A(t)cos(2f0+ ), in additive Gaussian noise. As the term noncoherent implies, the algorithm is capable of detecting signals with unknown phase, . Applied to ultrasonic NDE, the noncoherent detector is capable of detecting a family of transients obtained by continuously varying the angle over the interval [0,2] in signal s(t) above. Since it is impossible to exactly specify what transients to expect after propagation and reflection of ultrasound, the design of family detectors seems attractive. Examples of transients with the same envelope, A(t), and centre frequency, f0, but different phase angles, =0° and =90°, are shown in Fig. 1. These angles correspond to the transient being either symmetric or anti-symmetric.
Temporal resolution of SSP vs NCD
The temporal resolution of the highly non-linear SSP technique has previously been found to be inadequate for resolving closely separated targets, (cf. [8]). This could seriously limit the possibilities to size small defects. Our experience with real ultrasonic data indicates that the NCD algorithm shows no such problems. To quantitatively compare the temporal resolution of the two algorithms both of them were used for processing the same set of signals, containing two echoes separated in time, see Fig. 6. An arbitrary separation between the echoes was achieved by generating the signals in the computer, based on measured transducer characteristics.
Experimental results
The SSP/CPC and NCD techniques were evaluated using ultrasonic signals acquired from various specimens made of coarse structure materials, here we present some examples obtained for cast stainless steel specimen. The data acquisition was performed using a computerized ultrasonic instrument with an amplitude resolution of 8 bits. The acquired B-scans were then transferred to a PC running MATLAB for off-line processing.
Normal incidence transducers were scanned on a flat cast stainless steel block with structure similar to that of material used for housing of circulation pump in PWRs. The block dimensions were 150x200x295 mm; it had 3 side-drilled holes ( 5, 8 and 10 mm, respectively) placed at depth 40 mm below the scanned surface. Below, we present some results obtained using two contact transducers from Panametrics: 2.25 MHz M106 ( 12.7mm), and 2.25 MHz V325-SU ( 9.5mm).
Conclusions
In the paper two algorithms, intended for signal enhancement in ultrasonic inspection of coarse structure materials, have been presented and compared. The first algorithm, referred to as Consecutive Polarity Coincidence, is a robust version of SSP that does not require detailed a priori knowledge of the frequency range utilised for processing. The CPC automatically detects the widest frequency band (within the transducer frequency range) where the ultrasonic pulse has features of a mathematical pulse. It is worth noting that polarity thresholding SSP is a special case of the CPC with thresholding. The second algorithm, referred to as Noncoherent Detector, makes use of noncoherent detector statistics concept to design a kind of filter matched to the combination of material and transducer at hand. By employing such filter, the frequency range best suited for a particular material can be automatically estimated and utilised for the inspection, without the need to employ a tailor-made transducer.