Vision based quality control of composite components: considerations about measurement accuracy

In this paper a methodology is discussed concerning the measurement of yarn’s angle of a reinforced polypropylene matrix used in the production of automotive components. The measurement method is based on a vision system and advanced processing of images in order to evaluate the geometrical parameters of interest; the accuracy of measurement is a mandatory requirement, in order to assess the simulation approach for thermoplastic process optimization. Many aspects influencing the whole accuracy of the method have been identified and their effect evaluated, of both geometrical and optical type, allowing to perform angle measurements of the fiber angle with a whole accuracy in the order of a few degrees. By this way both local and extended defects can be identified in a reliable way also with reference to components of complex geometry. According to these results, accurate measurements of angle allows us to both validate the simulation of the thermoplastic process and to give suggestions for process improvement of fiber glass components of complex geometry.


Introduction
During the last decades, the use of advanced composites, in particular fiber reinforced plastic materials, has become more and more popular in a wide range of domains of activity, in particular in automotive and aerospace industries, due to their important characteristics, such as high strength-to-weight ratio, high stiffness-to-weight ratio, low density, wear resistance, chemical environment stability and long fatigue life, recyclability and their interesting capacity to be formed and produced in high volume rate [1]. Of great interest are the structural characteristics of the material, which influence the mechanical characteristics of the produced composite parts. In fact, fiber reinforced plastic materials, consisting of a polymer matrix reinforced by fibers, exhibit an orthotropic behavior due to fibers orientation. For this reason, it is crucial to consider the fibers orientation as a major factor in the entire technological process, starting from the design and finishing with components cutting [2, 3]. The fiber misalignment is a dominant defect of the tensile and compressive properties of these materials [4]. The textile draping process is one of the most critical steps in the production of fiber reinforced thermoplastic composite, from this point of view. In fact, in the draping process a flat textile has to be adapted to a 3D geometry, and this induces shear deformations and changes the local yarn orientation, which can be critical to the performance of the material [3]. These phenomena are usually being addressed experimentally by performing large deformation mechanical tests such as the bias-extension and the rotational bending which are complicated and conducted in the temperatures of reference in which each material can be thermoformed. Usually, the output of these tests is the basis of a process simulation using Finite Element (FE) modelling as indicated in [5][6][7]. Simulation plays a fundamental role as for process optimization, even though many aspects should be taken into account, with reference to both physical and chemical characteristics of materials and real operating condition during the production process [8][9][10], and integration with experimental evaluations is an unavoidable aspect of the approaches aiming at the optimization of the production process. To set up a simple validation system for the experimental assessment of the FE process, requirements concerning accuracy of measurements should be satisfied, in order to guarantee an effective interaction between simulation and experiments [11][12][13]. This goal is not simple, considering the many aspects interfering with measurements, which are multiple and of different type, related to the procedure, to the measurement system, to the material characteristics, only to cite the main ones. Different methods are proposed in literature for the measurement of yarn orientation in draped fabrics. Some methods measure the yarn orientation directly from digital images of the draped fabric by the use of image analysis algorithms such as edge detection or gradient methods [4,14]. These direct optical methods are useful for one-layer fabrics, or to inspect only the uppermost layer in multi-layer materials. A different method is based on the analysis of the bi-directional reflectance distribution function (BRDF) [15], which determines the yarn orientation by using the direction-dependent light reflection characteristics of the carbon and glass fibers. Image analysis is based on photometric stereo and uses different illumination patterns in combination with a fiber reflection model. The restriction of the analysis to the uppermost layer can be overcome by using electrical inspection methods. High-frequency eddy current testing is often used for the detection of fabric defects such as gaps and foreign materials [16][17][18], but it can be also used to visualize yarns in fabrics, inducing a circular eddy current in the specimen and measuring the resulting electromagnetic field effects [3]. For this method, the fibers of the composite material have to be conductive, which is the case for carbon fiber fabrics, but not, for example, for glass fiber composites. An interesting and promising line of research concerns the analysis of images acquired with polarized cameras. It is known, in fact, that the fiber strands influence the incident light and mainly reflect light of a certain polarization orientation. With a simple unpolarized illumination, the polarization of the reflected light can be measured and directly gives the orientation of the fibers, for the uppermost layer of fibers, even with resin being applied [19]. A variety of polarization sensors have been developed in the last years, among which camera-based polarization sensors, which can measure multiple direction information of polarization pattern [20,21] and which can be very useful in studying the deformations of the uppermost layer of fibers of composite materials. One important point that appears in research works where a comparison is made with simulation, is that a deviation of 10° or more between predicted and measured fiber or shear angles is generally found [3, 5, 22], then further research on both simulation and measurement methods is of current interest. In this paper a contribution to the set-up and optimization of the fabrication process of glass fiber thermoplastic composites is provided, by means of fiber angles measurements, carried out with more types of processing techniques of the acquired images. The methods will be critically evaluated and compared, to be used, in the future work, to validate simulation results. Indications from the uncertainty evaluation will allow to optimize the experimental set-up and the procedure in order to obtain the validation and the assessment of the simulation results at a suitable level of uncertainty for an effective control of the production process of parts made of thermoplastic composite material.

Materials and methods
The material under analysis consists of a polypropylene matrix reinforced with a woven 2-2 twill Eglass textile ( figure 1.a). The fiber volumetric content is 47%, while the melting point of the material in total is 163 °C. Nevertheless, the temperature's window opens approximately at 130 °C. Measurements have been carried out first on reference image (drawing made with a CAD software), to evaluate the accuracy of the methods, and then on a flat piece of the composite material analyzed, to evaluate the variability of the yarn angles on the planar starting plate. In the next phase, measurements have been carried out on a 3D object, obtained by thermoforming process, that is a component destined to support the battery tray of a car ( figure 1.b), a fact which signifies the importance of its integrity while in service [11]. To this end, extreme fiber deformation and variation of the thickness of the component may lead to undesired structural instability. A FLIR Camera, 2448x2048 pixels, has been used for the acquisition of the images. As far illumination, two soft-boxes with 80W lamp bulbs are used. In the next phase, measurements have been carried out on a 3D object, obtained by thermoforming process

Yarns angle measurement methods
The images are digitally post-processed in order to measure the angle between fibers; the processing is carried out by means of a specific, high performance software for geometrical analysis of elements in pictures, the Vision Builder for Automated Inspection, by National Instruments. Among the functions available in the Vision Builder software, those for edge detection are useful in the specific application. To locate an edge, a search direction has to be set (figure 2.a), along which the intensity differences between pixels are detected.  In this way the angle measurement is made using more spaced points, which is an advantage from the point of view of the accuracy of the angle measurement. However, in this way, strongly localized effects, if any, are attenuated. After a preliminary validation of both methods on a reference twill fabric picture, the following plan of the experiments has been carried out: a. Measurements on a flat plate whose nominal angles are 90°: -32 repeated measurements, realized with reference to the same couple of yarns.  Then, as would be expected, Method 2, which uses more spaced points for the measurement, presents a lower variability than Method 1. It can also be observed that the variability evaluated on the same pair of yarns takes into account the variability of the methods, which are affected by the way in which the regions of interest or the points used for the measurement are selected. Instead, the variability evaluated on different couples of yarns is also affected by the variability of the measurand, that is the superficial inhomogeneity typical of the analysed material; a rough estimate of the angle variability, based on the above results of both methods is in the order of 2-3 °. It must be observed that Method 2, although characterized by less variability, is less suitable for measuring angles on non-flat surfaces, where the deformation of the texture can undergo rapid variations passing from one point to another on the surface, because in these cases a more punctual measurement is required. For this reason, Method 1 has been chosen for the measurements on the 3D part. To more accurately determine the overall variability for Method 1, which integrates the contributions due to the method and

Measurements on the 3D object
Measurements have been carried out on the 3D object, shown in figure 1.b, obtained by a thermoforming process. The component was obtained starting from a flat sheet which, during thermoforming, undergoes deformations in the fabric due to the friction between sheet and mold. These deformations produce a concavity of the horizontal fibres, facing up or down in specific areas of the component, called zone A and zone B, respectively (figure 7). These defects are favoured by the fact that the matrix is made viscous by the high temperature of the furnace before forming, which allows a freedom of movement to the fibres, thus facilitating the deformation of the fibres themselves. To evaluate this effect, a reference system has been set, with y-axis in the direction of the lateral vertical fibres, and the α angle between the fibres and the x-axis (figure 8) has been measured in correspondence with 15 lines over the entire height of the lateral part of the object (figure 7). The aim is to define a reference system integral with the piece, independent of its orientation within the image. The definition of a reference system is not trivial and it influences in a remarkable way the whole uncertainty of the measured angles. Considering the zone A, in which the fibres have concavity upwards, it is clear that there must be angles close to zero in the central part of the zone, angles less than zero on the left side, and greater than zero on the right side (figure 8). For zone B, the trend is opposite, i.e. α is greater than zero on the left, equal to zero in the centre, and less than zero on the right. These theoretical observations are confirmed by the experimental results, which are represented as a heat-map in figure 9. These measurements can be useful information for the optimization of the process, since the simulation alone cannot easily predict these effects of localized deformation of the texture. Furthermore, resolution and uncertainty of results are low enough to suggest process modifications and improvements, in order to reduce unwanted grid distortions.

Conclusions
A methodology for the measurement of the yarn angle of a composite material of glass fibre has been discussed, based on a vision system and an advanced image processing procedure for process simulation validation. The method has proved to be able to identify local behaviour of fibre but also to describe surface trends of angle between fibres, where the acknowledgement of reduced variations of fibre orientation is able to suggest improvements to the thermoplastic process. This ability has been reached thanks to the satisfactory uncertainty of angle measurements based on image processing. The assessment of uncertainty required to identify the main aspects influencing uncertainty and to evaluate their contribution.  In particular, the main aspects to be considered are both geometrical and optical, like area of interest, yarn fibre boundary identification, camera resolution, lighting conditions and, in case of threedimensional objects and defects, also the local reference system. Calibration of the whole measurement system, with reference to this specific surface, is also of concern. Angle measurement uncertainty in the order of a few degrees allowed us to describe the studied defects in a satisfactory way; the future work will be devoted to check the reproducibility of procedure in different applications, also with very challenging geometries.