Elsevier

Additive Manufacturing

Volume 21, May 2018, Pages 598-604
Additive Manufacturing

Acoustic emission for in situ quality monitoring in additive manufacturing using spectral convolutional neural networks

https://doi.org/10.1016/j.addma.2017.11.012Get rights and content

Abstract

Additive manufacturing, also known as 3D printing, is a new technology that obliterates the geometrical limits of the produced workpieces and promises low running costs as compared to traditional manufacturing methods. Hence, additive manufacturing technology has high expectations in industry. Unfortunately, the lack of a proper quality monitoring prohibits the penetration of this technology into an extensive practice. This work investigates the feasibility of using acoustic emission for quality monitoring and combines a sensitive acoustic emission sensor with machine learning. The acoustic signals were recorded using a fiber Bragg grating sensor during the powder bed additive manufacturing process in a commercially available selective laser melting machine. The process parameters were intentionally tuned to invoke different processing regimes that lead to the formation of different types and concentrations of pores (1.42 ± 0.85 %, 0.3 ± 0.18 % and 0.07 ± 0.02 %) inside the workpiece. According to this poor, medium and high part qualities were defined. The acoustic signals collected during processing were grouped accordingly and divided into two separate datasets; one for the training and one for the testing. The acoustic features were the relative energies of the narrow frequency bands of the wavelet packet transform, extracted from all the signals. The classifier, based on spectral convolutional neural network, was trained to differentiate the acoustic features of dissimilar quality. The confidence in classifications varies between 83 and 89 %. In view of the narrow range of porosity, the results can be considered as promising and they showed the feasibility of the quality monitoring using acoustic emission with the sub-layer spatial resolution.

Introduction

Selective laser melting (SLM) is a powder bed AM technology, which allows building components with complex 3D geometries from an alloy powder layer by layer. Other terms that describe the same technology are Laser Cusing, Direct Metal Laser Melting, or Laser Metal Fusing. The technology has been successfully applied for rapid prototyping of unique workpieces with highly complex geometries that are impossible to produce using traditional forming methods [1]. It also has clear advantages in small series and individualized products for medical [2,3,4], turbine [5,6], aerospace [7,8], robotics [9], automotive and machine tool industries [10].

Even though a lot of progress has been made on behalf of the machine manufacturers to improve the process efficiency, the number of materials that can be reliably processed is currently still very limited (e.g. austenitic stainless steel, eutectic Al-Si alloys). This is the result of the extremely complex physical phenomena during the SLM process. It implies the very rapid consolidation of the base material powder in a small material volume (∼0.001 mm3) using a focused laser followed by cyclic heating and cooling at high rates (103–107 K/s) [11]. The processing parameters like laser energy, scan velocity, hatch distance, powder layer thickness, scan strategy, etc. have to be carefully adjusted to the alloy of interest. An improper set of parameters can lead to pronounced porosity, cracking and/or accumulation of residual stress inside the workpiece [12], resulting finally in poor mechanical properties [13,14]. The process repeatability is thus still limited, preventing the technology from being used in a much wider range. A possible solution to overcome this situation is the development of an in situ and real-time monitoring of the workpiece quality and to better control the machine operation [12].

At present, the industrial standard for monitoring the workpiece quality with regard to porosity or cracking is X-ray tomography [15]. The tests are carried out post mortem, when the machine time and materials are already spent. This method is also known for being very expensive and time-consuming. Consequently, online control is of a great demand and several approaches are seen to solve this problem. Temperature measurements of the processed zone with 1D or 2D pyrometers are used in multiple works [12,16,17]. The main objective is to keep the temperature parameters within a certain range that guarantees the mechanical properties of the workpiece to stay in an acceptable range. This approach uses built-in models for data analysis and provides a high spatial resolution quality control. Unfortunately, in real-life conditions, the multiple non-uniformities of the laser-matter interactions are responsible for a divergence between the models and the real-life situation resulting in performance inaccuracies. Image processing is another approach that is already introduced in a number of the available AM machines [12,18,19]. Quality monitoring is carried out with the visual inspection of each layer and an image processing routine searches for defects. The technical realization of such system is performed using matrix photodetectors in the visible spectral range. Again, one major weakness of this system is that the quality is monitored after an entire layer is produced. Besides, this method has certain limitations with regard to the spatial resolution and requires expensive optical systems to detect small defects.

The present work is a feasibility study that focuses on an alternative approach, in which acoustic emission (AE) is taken for quality monitoring of the AM process. In particular, the aim was to identify the laser processing regimes that potentially induce different types (e.g. due to incomplete particle melting or due to evaporation) and concentrations of porosity. The attractiveness of AE signals as compared to other methods lies in the high sensitivity of existing AE sensors and the relatively cheap hardware. The processing of AE signals is fast as it uses 1D data as compared to imaging (2D data) or tomography (3D data). Finally, AE sensors provide a high temporal resolution allowing to precisely localize the defects. The applicability of AE to detect defects during AM was already shown by Wu et al. [20] for polymer materials, although no in situ and real-time monitoring was proposed. Fiber Bragg gratings (FBG) are often used for detecting AE signals because of their high sensitivity in a wide acoustic spectral range [21]. The combination of high sensitive AE sensors and machine learning (ML) has been already successfully applied to a number of applications, such as tribology [22,23] and fracture mechanics [24]. All these works are characterized by complexity in acquired data structure and noisy environments, thus making this combination promising for AM, as well.

In recent years, convolution neural networks (CNN) has become a popular technique for acoustic tasks [25] showing high efficiency and possibilities to suppress the stationary noises. Nevertheless, the conventional CNNs have certain limits when analyzing non-regular data. Spectral convolutional neural networks (SCNN) are a recent extent of conventional CNN with improved efficiency in classification/regression tasks that are operating on irregular data grids [26,27,28]. High performance of SCNN is achieved combining the best from: i) deep learning, inherited from conventional CNNs, and ii) spectral graph theory [26]. In the latter case, the local spatial structure of the input data is recovered using a weighted graph G. The learning of the local irregularities in SCNN is carried out using the convolution operator, which is realized by applying G upon the input signals f. In conventional CNN, the spatial convolution using regular kernels fails when operating on the same type of data [26,27,28]. Currently, the SCNN methods are mainly reported for image processing [26,27,28]. However, the representation of acoustic signals in the time-frequency domains using wavelets allows applying already developed 2D image processing methods, making SCNN a ready “on-shelf” solution for acoustic analysis.

The modelling of the data irregularities in SCNN is possible using spatial and spectral approaches, and both exhibit the same classification efficiency [29]. However, the spectral approach was chosen in this work as the spatial approach does not provide the time ordering [29], which can be valuable information for the localization of the individual defects in the future. More information about the SCNN can be found in the two comprehensive reviews [29,30].

In this study, the AE signals were collected during the SLM of a stainless steel using a fiber Bragg grating. Different laser processing parameters resulting in different part qualities with regard to the amount of porosity were selected. The AE signals were evaluated and the classification of the process in terms of “high”, “medium” and “poor” quality was tested using both, conventional CNNs and SCNNs.

Section snippets

Data acquisition setup

The SLM process of CL20ES stainless steel (1.4404/316L) powder (particle size: 10–45 μm) was performed on an industrial machine Concept M2 (Concept Laser GmbH, Germany). The machine is equipped with a fiber laser operating in continuous mode at a wavelength of 1071 nm with a beam quality M2 = 1.02 and a spot size of 90 μm. The Concept M2 was additionally equipped with one fiber Bragg grating (FBG) sensor to detect the airborne AE signals, generated during the AM process.

The FBG sensor is an

Features extraction

In the present work, the relative energies of the narrow frequency bands were taken as the input features for the SCNN classifier. The frequency bands were extracted using a standard wavelet packet transform (WPT) [35]. WPT is an extension of the traditional wavelet transform that can be represented as a pass of the signal f through a set of filters [35]:φj(n)=nh0(k)Mφ(Mnk),kZψji(n)=nhm1(k)Mψ(Mnk),kZwhere ho is a low pass and hm high pass filters, φ() and ψ() are the scale and wavelet

Collected datasets

The collected AE signals were divided into three categories according to the manufacturing quality of the workpiece layers described in Section 2.2 and are shown in Fig. 2. A number of patterns, bounded by the RW, were collected from the AE recorded signals to form two datasets: one for training and one for the test. Each category (that corresponded to poor, medium and high quality) in each dataset was equally represented by 300 patterns with no common RW between both datasets. This approach

Conclusions and future work

This work presents the results of a feasibility study for in situ and real-time quality monitoring using acoustic emission (AE) and machine learning (ML). A fiber Bragg grating (FBG) sensor, mounted directly inside the process chamber 20 cm away from the process zone, provided a high sensitivity in signal detection. The airborne acoustic emission (AE) signals were detected and further grouped in terms of AM processing quality. The Daubechies wavelet with ten vanishing moments was used to

References (40)

  • J. Scott et al.

    Additive manufacturing: status and opportunities

    Report from Sci. and Techn. Policy Insitute (IDA)

    (2012)
  • D.A. Türk et al.

    Additive manufacturing with composites for integrated aircraft structures

  • C. Semini et al.

    Additive manufacturing for agile legged robots with hydraulic actuation

    Int. Conf. Adv. Rob. ICAR

    (2015)
  • L.C. Zhang et al.

    Selective laser melting of titanium alloys and titanium matrix composites for biomedical applications: a review

    Adv. Eng. Mater.

    (2016)
  • K.S. Everton et al.

    Review of in-situ process monitoring and in-situ metrology for metal additive manufacturing

    Mater. Des.

    (2016)
  • J.J. Lewandowski et al.

    Metal additive manufacturing: a review of mechanical properties

    Annu. Rev. Mater. Res.

    (2016)
  • S.F. Shirazi et al.

    A review on powder-based additive manufacturing for tissue engineering: selective laser sintering and inkjet 3D printing

    Sci. Technol. Adv. Mater.

    (2015)
  • A. Thompson et al.

    X-ray computed tomography for additive manufacturing: a review

    Meas. Sci. Technol.

    (2016)
  • W. Devesse et al.

    Hardware in the loop control for additive manufacturing processes using temperature feedback

    J. Laser Appl.

    (2016)
  • M.H. Farshidianfar et al.

    Real-time control of microstructure in laser additive manufacturing

    Int. J. Adv. Manuf. Technol.

    (2016)
  • Cited by (311)

    View all citing articles on Scopus
    View full text