Optimizing the automation of an iron ore production line — A case study, part II: Optimal automated quality control

The preprocessing part of an iron ore mining production line is being modelled, together with description of the algorithms for its optimization with respect to criteria and constraints corresponding to the relevant industrial standards. We use as a case study the production line of the international high-tech mining and processing company LKAB stationed in northern Sweden. The modelling is generic, i.e., the exposition constitutes a general description of the types of constraints, criteria and mathematical algorithms without providing essential technological details which are intellectual property of LKAB. The exposition is divided into two inter-related articles, of which this is the second one, the first one being [1]. This second article is dedicated to optimization of the automatic quality control of the output from the preprocessing part of the production line which serves as input to the first phase of the processing part: the refinement phase (for the conceptual structuring of the production line into consecutive parts, phases and stages we refer to [1]). Quality is being evaluated based on the graphical analysis of images output by automatic cameras installed: 1. at the so-called `alternative obligatory points' (AOPs) [1] on the transportation routes in the mine - in the preprocessing part of the production line; 2. above the conveyor belt containing `green' pellets - at the `cold' stage of the pelletization phase of the processing part of the production line. In this article, two different types of graphical analysis are considered, as follows: A. based on a 2D-planar image-processing algorithm; B. based on a 3D-stereo image-processing algorithm. We discuss the main blocks in the flowcharts of these algorithms and compare their performance. Each one of them is shown to have advantages and disadvantages of their own in comparison with the other one; for best results we recommend the combined use of both, and outline several ways in which the benefits of their combined use can be maximized. The study is concluded by discussing how to tune the respective criterion-generating blocks of algorithms A and B (in the case of separate use of these algorithms, as well as in the case of combined use of both algorithms) in order to achieve optimal performance on images of type 1, resp. type 2.