You are on page 1of 6

New Analytic Approach for Acoustic Material Testing

Dipl. Wirtsch.-Ing. Jörg Ritter


RTE Akustik + Prüftechnik GmbH, Germany, j.ritter@rte.de
More info about this article: https://www.ndt.net/?id=28299

Abstract

Acoustic material testing is well known for quality inspection of metal and composite
parts in serial production, with short cycle times and a robust performance. Acoustic
material testing is a referencing respectively comparative testing method. Therefore, pre-
classified reference parts are required to teach the testing system. In practice, it is often
difficult to provide clearly pre-classified failure parts. Also, it is common to use sample
parts which are characterized by the subjective evaluation of the employees. In this case
evaluation deviations between the subjective evaluation and the objective measurement
result of the acoustic test system are to be considered. This leads to discussions, as the
employees tend to trust their own judgment and rely on their experience. Defects must
then be clearly determined, by using other and more complex and costly testing methods,
such as ultrasound or CT. With new analytic techniques the teaching process, especially
without the availability of failure parts, can be shortened and simplified.

KEYWORDS: Acoustic Material Testing, Acoustic Resonance Testing, FRF

1. Introduction
Acoustic material testing [1] is a referencing as well as a comparative testing method.
Pre-classified reference parts are required to teach the testing algorithm. It is often
difficult to provide clearly pre-classified defective parts or samples with natural defects.
It is common to use objective classified samples as reference parts for the teaching of the
acoustic testing system. Defects must be clearly determined by using other and more
complex and costly testing methods, such as ultrasound, CT scanner or even destructive
testing methods. The teaching process of an acoustic material testing system (ART) for
suitable test criteria, must be done iterative by re-assessment for improving the testing
characteristics. Common acoustic testing systems are using conventional characteristics
like frequency position, frequency damping, amplitude monitoring or amplitude
attenuation. But this is very often not sufficient for a reliable testing, in consideration of
batch or production tolerances. Modern ART systems offer a good selection of intelligent
algorithms in combination with characteristics based on pattern recognition, like
frequency split up, frequency distance relations, multiple frequency peak detection,
frequency compensation and more. An additional approach is the evaluation of frequency

© 2023 The Authors. Published by NDT.net under License CC-BY-4.0 https://creativecommons.org/licenses/by/4.0/


bands instead of just individual frequencies. It becomes the fundament of advanced
methods of acoustic material testing.

2. The practical approach


Acoustic material testing [1] utilises the physical effect of a component stimulated to
vibrate in a reproducible manner in its natural frequencies. Anomalies in the frequency
response function (FRF) under identical conditions, allow conclusions to be drawn
regarding the changes in the material structure and its strength. The anomalies are caused
by the deviation in the material structure of the test parts.
For the practical approach, the subjectively assessed test parts are measured using the
ART test system. The obtained differences of the objective and subjective results must be
analysed in detail. Some of the test parts subjectively assessed as "good" (OK) were rate
by the test system as suspicious (faulty, NOK) and some of the test parts subjectively
assessed as "faulty" are assessed by the system as "good".
In the selected practical example, the components with a faulty material structure can
only be explicitly verified through destruction. Test samples had to be used for the trial
being subjectively classified in advance. Enough non-classified test parts were
subjectively and objectively tested. The objective assessment was carried out using
standard frequency test criteria and compared to the results of the subjective test. The
measurement results of the test system were reproducible and could thus be accepted as
fact.
The following figure (Figure 1) clearly illustrates the deviations between the objective
and subjective assessment using a typical test criteria range. The red dots represent all
NOK evaluation (subjective and objective).

Figure 1: Illustration of the measurement results including the subjective assessment.


(X-axes shows no. of samples, Y-axes shows characteristic value (frequency variation mV))

The two red lines in Figure 1 represent the test criteria range, within the measurement
results of the "good" and fault-free (OK) parts are expected. The way in which red and
green measurement points are mixed up in the good range is clearly visible.
Figure 2 illustrate a random sampling of 2091 test parts with 91,1% similar evaluation by
results from subjective and objective assessment. While 8,9% of test parts show an
inconsistent rating and demand a follow-up investigation.

2
Figure 2: Statistical analysis of the subjective and objective assessments

The difference between the subjective and objective classification in the selected example
was further complicated by the fact, that the test parts were produced with different
manufacturing methods (two different press methods for material compacting). As a
result of the different manufacturing processes, the samples show differing natural
frequency behaviours depending on the press method. This variation effect was
significantly increasing the complexity of the testing task.
Figure 3 shows the deviating measurement results through different manufacturing
processes, by identical component types being produced. For Process 1, a double peak
can be seen in the spectrum (frequency range 3.800 to 6.000 Hz), while Process 2 result
in just one frequency peak in the same frequency band.

Figure 3: Measurement results of manufacturing process 1 (left) and process 2 (right),


all good parts in the range 3.800 to 6.000 Hz
(X-axes shows frequency, Y-axes shows amplitude mV)

If you now superimpose the measurement results of the test parts from both methods that
were subjectively rated as good and subjectively rated as faulty in one spectrum, it
becomes apparent that the problem arises of designing a sole and reliable test
characteristic.
Illustrating this, using conventional approaches is an elaborate process. This results in an
initial solution approach, up to twenty-four common acoustic characteristics
(Figure 4) over the complete resonance spectrum, to clearly classify reliable the NOK
samples. Common characteristics were chosen in the frequency domain: frequency shift,
frequency split, amplitude monitoring, and amplitude attenuation.

3
Figure 4: Statistical analysis of the individual characteristics
(X-axes shows no. of characteristic, Y-axes shows no. of samples)

Such a large number of test criteria results in a prominent level of complexity of test
criteria. A knowledge-intensive training of quality engineers is required. The customer
therefore requested a significantly reduction of the number of test criteria to only one
characteristic, applicable for the different production processes, without having to reduce
the accuracy rate to detect faulty parts.
A new analytical approach was considered to monitor specified frequency ranges of the
resonant spectrum, instead of looking after individual frequencies anomalies. This new
approach for acoustic resonance testing, considering evaluation methods from NVH
testing (Noise, Vibration and Harshness), like cumulative or average level, quadratic sum,
RMS or envelope detector, standardly carried out on testing electrical drives. NVH
evaluation methods are applied mainly in the time domain of a recorded time signal.
Acoustic resonance methods are mainly applicable in the frequency domain (spectrum).

The considered approach using NVH methods like windowing, band level energy,
quadratic sum, RMS and its difference. By using a frequency band level characteristic
(Fig. 5), the influences of the different production processes could be separated from
measurement results of the faulty parts. Similar frequency band level evaluation was
chosen detecting the NOK parts (yellow window in Figure 5).

Figure 5: New criterion with monitoring of several band ranges


(X-axes shows frequency, Y-axes shows amplitude mV)

4
However, it still requires a manual process to develop the right frequency band with its
anomalies to configurate the testing characteristics. In our approach the test system is
configurated that way, that all results which do not correlate with the expected spectrum
pattern of OK parts, are assessed as "suspicious" and will be sorted out. The NOK
components which are sorted out, are double checked using other suitable test methods.
The test system can thus be continuously optimised, and the pseudo rejection rate can be
reduced (Figure 6).

Figure 6: Statistical analysis of the results of the new evaluation approach


(Sum-spectra variance of band 3.800 to 4.200 Hz)
(X-axes shows no. samples, Y-axes shows characteristic value)

The described manual process of re-assessment and optimisation of the test criteria to
improve the accuracy rate should also be automated in a subsequent step.
Automatic tracking of the test criteria is a known option to simplify the process for users.
The limit curves of a criterion are consistently tracked and changed [2].
The idea is, to emulate process fluctuations resulting from production variations and
environmental influences to minimise the rejection rate. A disadvantage of this process is
that the fluctuations are not assessed as errors or process faults and, depending on the
algorithm, error effects could be masked. However, the target value and limits of many
factory standards are not allowed to be changed, because, for example, existing reference
parts might no longer be recognised.
Another step toward to a self-learning test system is taken with the aid of mathematical
pattern recognition processes and automatic re-assessment feedback. Initial trials in
collaboration with the IDMT Fraunhofer Institute demonstrate great successes and
promise users significantly simpler implementation and handling of acoustic material
testing systems.
The familiar statistical teaching-in processes, calculate the average from a batch sample
of test parts. For example, for selected simple criteria using a statistical approach and the
limits are set with a specified tolerance (e.g. +/- 3%). However, the criteria and the limits
are still manually determined by the user as before. The use of support vector machines
seems to be a promising method for an automatic teach-in process.
In contrast, the new approach with frequency band level criteria, provides consistent
criteria, considering the process fluctuations and always allow for the correct detection of
the selected reference parts.

5
3. Conclusions
The innovative aspect of the new analytic approach is the use of specific NVH
characteristics, based on the methods for noise testing and the application of pattern
recognition techniques, which leads to the future implementation of artificial intelligence
technologies. With these innovative approaches, the acoustic testing will become more
usable for all metal and hybrid materials with more acceptance for quality inspection.

References
[1] Dipl.-Inform. Ingolf Hertlin, “Acoustic Resonance Analysis”, NDT Compact and
understandable, Castell 2003
[2] Dipl.- Dagmar Metzger, “Überwachung von Qualitätsmerkmalen bei Komponenten
und Bremssystemen“, VDMA 2020

You might also like