Characterisation of High Power Electron Beams using a Two-Slit Probe and Wavelet Transforms
Aman Kaur
TWI and Brunel University
Colin Ribton TWI Ltd, Granta Park Great Abington, Cambridge, CB21 6AL, UK
Professor W. Balachandran Brunel University
Paper presented at Intelligent Signal Processing Conference. ISP 2015, 1-2 December 2015, London, UK.
Abstract
High power electron beams are used for welding critical components in aerospace and nuclear industries due to their inherent advantages. There are high quality requirements in these industries and hence the associated cost of materials and processes is also very high. This makes it very important to ensure that the beam quality is maintained and checked prior to carrying out the welds. The processes in these industries are highly controlled, however, even minor changes in the operating parameters of the electron beams generated by electron guns can make large enough variations in the beam quality that can result in poor welds. Many devices and techniques exist to measure the beam quality, however, these are limited in their operation at high powers. A two-slit probing system has been designed to measure the beam quality that can be used at high powers. The device consists of two slits in the x and y axes which collects a small portion of the beam current when the beam is deflected over the slits. The signals received from the device are processed using wavelet transforms to differentiate between the signals within the design space and out of it. The starting point is the assumption that different quality of beams will possess unique signal profiles in the form of the distribution of energies with respect to frequency and time. This work applies wavelet decomposition to beam probe signals to develop a unique method that finely determines variation in beam quality.
1. Introduction
The advantages of Electron Beam Welding (EBW) over other methods i.e. its ability to produce very narrow welds with deep penetrations, narrow heat-affected zones and minimum distortion, makes it useful in applications where high precision is required. Over the years, the requirements for aerospace and nuclear industries have resulted in stringent quality specifications that can be achieved by EBW. To carry out the process, high power electron beams are generated by electron beam guns which are focused on the work piece to be joined. The quality of the generated beam depends on various parameters like accelerating voltage, beam current, focus settings, vacuum levels in the chamber and in the gun, working distance and welding speed [1]. Usually, the accelerating voltage of the welding machines is fixed and the power of the beam is controlled by varying the beam current. Once other parameters are fixed for carrying out a weldment, the focus settings are usually set by the operator by visual methods and based on their experience which can vary from operator to operator [2]. There could be ±20% to ±40% variations in the weld depth just due to manual focus adjustment by different operators [3]. Also, there could be variations in the beam quality due to uncontrolled parameters such as noise.
It has been seen that even minor variations in the focus settings or variations due to noise can result in large enough variations in the beam quality and can result in the poor welds. This is unacceptable in the high precision applications and can cause scrap of materials and loss of productivity. This in turn demands a measurement system that can measure the beam quality before carrying out the welds. There are many devices in the literature and commercially available for welding applications. However, their operation is limited to low powers. A brief comparison of their relative merits and limitations has been mentioned in [4]. These devices have been used to correlate the weld dimensions with beam energy distribution and beam diameters. In this research work, a two-slit probing system has been employed that can be easily used up to 40kW of power. This device has been used previously to study the focus sweep in relation with the weld dimensions.
In the present work, it has been attempted to derive a design space in terms of parameters of the beam traces obtained from the probing system to detect the point where weld quality measures approach a threshold derived from the aerospace standards. To achieve this, the possibility of using the wavelet analysis to create the features vectors along with peak current intensity, full width half maximum beam width has been studied.
2. Similar work
Wavelet transforms have been used in the field of signal processing due to their ability to provide signal information in both the time and frequency domains. There are different functions or waves that can be used for wavelet transformation known as mother wavelets. Another advantage of using the wavelet transform over other techniques is that the mother wavelet can be selected based on the signal characteristics and the application. There has been intensive research carried out in different application areas such as medical, fault classification in power systems, acoustics, ultrasonics, image processing and finance to use wavelet transforms for feature extraction and classification [5,6]. A limited amount of research has been found on the use of wavelet analysis in electron beam welding and mainly this has been concerned with detection of secondary current signals using X-rays or ion detectors. Also, these applications are limited to aiming to provide a broad classification of the signals as compared to the present study where the derivation of a narrow box of parameters for a set of weld quality parameters has been developed.
Yoon in [7] has analysed the signals of X-ray and ion detectors using Fourier analysis and also using wavelet transform. He could differentiate between partially penetrated, fully penetrated and over-penetrated conditions better with wavelets as compared to using Fourier analysis. In [8], wavelet analysis has been used to identify the frequency range of the maximum energy of the beam based on Root Mean Square (RMS) deviation of the Discrete Wavelet Transform (DWT) coefficients. As the different decomposition levels possess different energy levels, the starting point of the present analysis is based on the assumption that beams of different quality will possess unique signal profiles in the form of the distribution of energies at different decomposition levels.
3. EB signal acquisition
The EB signals were acquired through the two-slit probe which consisted of two probe fingers mounted perpendicular to each other and a Faraday cup. The two probe fingers were used to capture the signal in x and y axes. The Faraday cup was used to measure the full beam current. The experimental set-up to acquire the signals is shown in Figure 1. Because of the high power of the beams, these were not continuously focused on the probe fingers, rather the beam was deflected over the fingers intermittently at very high speed to capture the measurement. The electron beam was deflected over the two-slit probe in the pattern shown in the red. When the beam travelled over the probe fingers, part of the beam current was collected by the Faraday cup of the probe finger which was then converted to a voltage signal through a resistor. This signal was captured either on a digital oscilloscope or through signal processing software on the computer.
Figure 1: Two-slit probe with scan pattern.
4. Processing of acquired signals
In this investigation, the probe traces from the two-slit probe were acquired at different beam current and focus settings. For the experimentation, 4 beam current and 5 focus levels shown in Table 1 were selected one-by-one, whereas the other parameters such as accelerating voltage, vacuum level, working distance and welding speed were kept constant. The focus settings were set around the centre focus which was adjusted for sharp focus at the lowest beam current value.
Table 1: Beam current and focus settings.
Focus Settings
Beam Current (in mA)
|
U2 |
U1 |
FC |
L1 |
L2 |
7.5 8.0 8.5 9.0 |
|
|
|
|
|
4.1. Wavelet transforms and feature extraction method
In the case of electron beam welding, penetration of the welds is related to the peak current intensity and the beam diameter, the Full Width Half Maximum (FWHM) pulse width, which were directly measured from the acquired probe traces. To acquire further parameters, wavelet transform techniques were performed on the acquired signals.
The discrete version of the wavelet transform (DWT) has been used to decompose the discrete signal into components under different scales of fixed wavelet functions called the “mother wavelets” [5, 9]. There are many functions like Harr, Daubechies (db), Coifllets (Coif), Symlets (Sym) that are used as mother wavelets. DWT acts as a pair of complementary high pass and low pass filters and iteratively decompose the signal into multi-resolution subsets of detail (D) and approximate (A) coefficients. The decomposition process of the DWT at different levels is shown in Figure 2. Each level is half the bandwidth of the one level above it.
Figure 2: DWT decomposition levels.
In the present analysis, the entire frequency range is decomposed into 8 levels using db3 mother wavelet. As it is important to choose a mother wavelet having a similarity with the signal to be analysed, various mother wavelets were examined for correlation coefficients [10] and percent root mean square difference [11]. The best results for the signals acquired from the slit probe were achieved using the db3 wavelet function. Also, the number of decomposition levels depends upon the sampling frequency of the acquired signal. In this case, the signal was captured with a sampling rate of 25MHz. Multi-signal analysis in 1-D has been carried out using proprietary software [12]. Figure 3 shows the decomposition levels of one of the probe signals.
Figure 3: Decomposition levels of the acquired signal
Table 2 presents the frequency ranges for the corresponding decomposition level.
Table 2. Frequency ranges for decomposition levels.
Decomposition level
|
Frequency band(in MHz)
|
1 2 3 4 5 6 7 8
|
6.25 – 12.5 3.12 – 6.25 1.56 – 3.12 0.78 – 1.56 0.39 – 0.78 0.19 – 0.39 0.10 – 0.19 0.05 – 0.10
|
The feature extraction is based on the parameters derived from the wavelet co-efficients at different decomposition levels. These include the total energy of the signal and the energy distribution among the detailed co-efficients and the approximate level among the decomposition levels.
Energy for each decomposition level is calculated as given in equation (1).
Where j is the decomposition level and n is the nth sample of the acquired signal. Also the normalised wavelet energy or the percentage of energy at different decomposition levels is given by equation (2).
Where Etot is the total energy of the signal.
In the present analysis, the set of parameters of the probe traces also called the feature vector, must contain both, the total energy as well as the percentage of energy. The total energy is contributing to the weld pool dimensions and the percentage of energy to represent the beam characteristics.
5. Results
The probe traces were captured for 20 signals at different beam current and focus settings mentioned in the previous section. Various features of the signals were captured to characterise the probe traces.
5.1. Features based on direct measurement of the signal
The peak intensity and FWHM were measured directly from the acquired signal and are tabulated in Table 3 and Table 4 respectively.
Table 3. Peak values of acquired signals.
Focus Settings
Beam Current (in mA)
|
U2 |
U1 |
FC |
L1 |
L2 |
7.5 8.0 8.5 9.0 |
3.5 3.76 3.73 4.01 |
4.85 4.97 5.13 5.45 |
5.9 6.09 6.01 6.37 |
5.77 5.88 6.31 6.55 |
4.08 4.44 5.14 4.88 |
Table 4. FWHM of acquired signals.
Focus Settings
Beam Current (in mA)
|
U2 |
U1 |
FC |
L1 |
L2 |
7.5 8.0 8.5 9.0 |
2.26 2.28 2.53 2.56 |
1.60 1.71 1.82 1.87 |
1.31 1.37 1.56 1.55 |
1.51 1.51 1.47 1.52 |
2.15 2.07 1.95 2.19 |
Figure 4 and Figure 5 represents the peak signal and FWHM respectively with respect to the different beam currents and the focus levels set around the sharp focus setting.
Figure 4: Peak signal for different beam currents and focus levels.
Figure 5: FWHM for different beam currents and focus levels.
5.2. Features based on wavelet transform of the signal
The acquired signals were processed using wavelet transform. The signals were decomposed into 8 levels. The total energy of the signals and their distribution among different decomposition levels were considered as the features vector as mentioned in the earlier paragraph. Figure 6 – 10 represents the total energy of the signals at different beam current and focus levels, the distribution of energy at d8, d7, d6 decomposition levels and a8 approximation level respectively.
Figure 6: Total energy of the signals.
Figure 7: Energy distribution at d8 decomposition level.
Figure 8: Energy distribution at d7 decomposition level.
Figure 9: Energy distribution at d6 decomposition level.
Figure 10: Energy distribution at a8 approximation level.
6. Discussion
The features extracted in the above paragraphs give a clear distinction of differing beam qualities at different input settings of beam current and focus levels. The peak signal values and FWHM indicated the trends as expected. At larger beam currents, the peak values were increasing and at sharp focus showing the maximum. Similarly, at sharp focus settings, the peak values were at maximum and decreasing either side i.e. at over-focused and under-focused levels. FWHM was narrowest at the sharp focus and becoming broader on either side. However, it was observed that the peak values at 8.5mA and 9.0mA were slightly shifted towards the lower focus levels i.e. L1. Also, the FWHM values representing the sharp focus were also shifted. From this, it was appearing that the sharp focus had shifted towards the lower focus level as the beam current was increased. This was in line with the explanation given in [13] i.e. the focus of the beam depends on the focus coil current as well as on the control-electrode voltage. Although the focus coil current for a typical setting is constant, however, the control electrode voltage changes with the beam current. Hence, the shift in the sharp focus at higher beam currents was observed.
The wavelet coefficients indicated that there were no high frequency components above about 400kHz. In a few signals only a small portion of the energy was observed. Other than the last approximation i.e. a8, rest of the energy was distributed among detailed levels d8, d7 and d6. The maximum change in the energy level was found in d7 around 76% from minimum to maximum value. The distribution of the energy percentages in these levels showed similar trends as that of peak values and FWHM i.e. the energies at these levels were increasing as these were moving towards the sharp focus as well as the levels were increasing with the increase in beam current. The lower levels of energy towards the sharp focus in distribution at approximation level a7 also indicated that as the beam was more focused, the low frequency contents decreased and higher frequency contents increased (i.e. the pulse shape becomes sharper). Also, the energy distribution among the detailed levels d8, d7 and d6 might be due to the pulse width that varies between 4 – 12µs and was represented by the frequency bands of these details levels.
The results of the work show that there is significant differences between the energy distributions among the decomposition levels. Even a 0.5mA difference in the beam current gives sufficient difference in the energy levels distribution. This suggests that the features distinguishing between different beam quality traces may dominantly lie in the total energy of the signal and their distribution among different decomposition levels. Through this work, it has been demonstrated that the wavelet transforms can potentially be used in characterising the probe traces to define features vector for the beam quality.
7. Conclusion
The present work has shown a novel application of wavelet transforms to allow distinction between finer levels of variation in beam quality. A new method was applied to analyse the signals obtained from the slit probe. The direct measurements from the signal and the features extracted from wavelet transform co-efficients were combined in order to distinguish between the different beam qualities. The further work on this will be involving to carry out the weldments and the probe traces simultaneously and to relate weld quality parameters with the beam traces’ feature vectors.
Acknowledgements
The present study has been supported by TWI, Cambridge and Brunel University, London as a part of a PhD programme.
References
- U. Dilthey, J. Weiser. “Investigations of EB characteristics and their influence on the weld shape”, Welding in the world, volume 39, pp. 89-98, (1997).
- J. Elmer. “Characterization of defocused electron beams and welds in stainless steel and refractory metals using the enhanced modified faraday cup diagnostic”, Lawrence Livermore national laboratory, pp. 1-9, (2009).
- W. Giedt, L. Tallerico. “Prediction of electron beam depth of penetration”, Welding journal, volume 67, pp. 299-305, (1988).
- A. Kaur, C. Ribton, W. Balachandaran. “Electron beam characterisation methods and devices for welding equipment”, Journal of materials processing technology, volume 221, pp. 225-232, (2015).
- A. M. Gargoom, N. Ertugrul, W. L. Soong. “Comparative study of using different mother wavelets on power quality monitoring”, Australasian universities power engineering conference, (2004).
- M. Sifuzzaman, M. R. Islam, M. Z. Ali. “Application of Wavelet transform and its advantages compared to Fourier transform”, Journal of physical sciences, volume 13, pp. 121-134, (2009).
- C. S. Yoon. “Electron beam welding diagnosis using wavelet transform”, Journal of KWS, volume 21, pp. 639-645, (2003)
- D. N. Trushnikov. “Using the wavelet analysis of secondary current signals for investigating and controlling electron beam welding”, Welding International, volume 27, pp. 460-465, (2013).
- R. J. E. Merry. “Wavelet theory and applications: A literature study”, Eindhoven, (2005).
- R. Umamaheswari, R. Sarathi. “Feature extraction of UHF PD signals by Wavelet packet based MRSD analysis”, IEEE, pp. 1218-1222, (2012).
- R. Khanam, S. N. Ahmad. “Selection of wavelets for evaluating SNR, PRD and CR of ECG signal”, International journal of engineering science and innovative technology, volume 2, issue 1, pp. 112-119, (2013).
- M. Misiti, Y, Misiti, G. Oppenheim, J. Poggi. “Wavelet toolbox: For use with Matlab”, The mathworks (2009).
- S. Schiller, U. Heisig, S. Panzer. “Electron beam technology”, John wiley and sons, (1982).