AUTOMATED RECOGNITION OF AUTOCLAVE SURFACE DEFECTS USING POINT CLOUD ANALYSIS

Abstract

The problem of detecting defects on the surface of autoclaves is extremely relevant for ensuring safety, quality, and economic efficiency of production in critical industries such as the food industry. Defects, including cracks, corrosion, and irregularities, create risks for sterility, product quality, and can lead to emergency situations. The article analyzes modern methods of scanning the internal surface of autoclaves, including laser scanning and photogrammetry, and identifies their advantages and disadvantages in terms of accuracy, speed, and cost. The aim of the research is to develop an effective automated algorithm for processing the results of scanning the internal surface of an autoclave to detect defects. The input data is a point cloud obtained by photogrammetry. The stages of building a dense point cloud based on Structure from Motion (SfM) and Multi-View Stereo (MVS) algorithms are described. The developed point cloud processing algorithm is presented, which includes sequential filtering of points based on the analysis of normal coordinates calculated using the Principal Component Analysis (PCA) method with the k-nearest neighbors algorithm. Five filtering stages aimed at extracting points that may correspond to defects are described: removal of isolated points and points with degenerate local geometry (zero normals), removal of points belonging to large planar surfaces and symmetrical elements (identical values of normal coordinates), filtering of points with a nearly horizontal orientation (z-coordinate of the normal close to one), selection of points located below the main surface (negative z-coordinate), and selection of points with normal coordinates deviating from the horizontal orientation by a threshold value. The results of applying the developed algorithm to a point cloud representing the internal surface of an autoclave with traces of corrosion are presented. Visualization of the point cloud after each filtering stage is demonstrated. The importance of comparing the results of the current scan with previous ones for monitoring the dynamics of defect development and evaluating the effectiveness of repair work is emphasized. The developed automated approach contributes to increasing the efficiency and objectivity of the process of detecting defects on the surface of autoclaves.