Artificial intelligence monitoring of direct strip casting

在生产线上识别有缺陷的产品是一个劳动密集型和成本高昂的过程，导致产品保持时间长和/或损失时间和精力。与质量有关的成本是一个重大的经济和环境负担：即使是生产工艺或材料(许多人肉眼看不见)的微小变化也可能导致完全有缺陷的生产运行。传统的基于机器视觉的表面检测方法对于边缘/对比度较强的缺陷检测效果较好，但受噪声和光照条件的影响较大。这项提案将解决一种新型铝带坯连铸设备中的缺陷检测问题。为了更好地处理噪音和照明问题，而不是传统的图像处理，该提案将开发最先进的机器学习方法，结合工业就绪的视觉捕获技术。该提案最初侧重于材料的事后评估，还将致力于开发一种在线缺陷检测方法。通过适当的开发，最终目标将不仅仅是检测缺陷，还将帮助指导给定铝合金的带钢连铸工艺的优化。该提案汇集了两个规模较小但具有创新性的行业贡献者，分别专注于机器学习和新型铝铸造技术的应用用途。虽然机器学习方法将在这里应用于工业铝带铸，但我们认为这是一个开发试验台，用于开发可直接转移到一系列连续生产行业的程序和过程。因此，为了能够在广泛的行业中应用，开发的方法将是模块化和灵活的。我们的目标是解决巨大的知识差距，并有可能使加拿大的各种行业受益。

Identifying defective products on a production line is a labour-intensive and costly process, resulting in long product hold times and/or lost time and effort. Quality-related costs are a significant economic and environmental burden: even a slight variation in production processes or materials (many invisible to the human eye) can result in an entirely defective production run. Conventional machine vision-based surface inspection methods work well for surface inspection when defects have strong edges/contrast, but they are greatly affected by noise and lighting conditions. This proposal will address detection of defects in a novel aluminum strip-casting facility. To better handle noise and lighting issues, rather than conventional image processing, this proposal will develop state-of-the-art machine learning approaches, coupled with industrial-ready visual capture technologies. Initially focused on postmortem assessment of material, the proposal will also work to develop an on-line defect detection methodology. With suitable development, the final goal will be not just detect defects but help guide optimisation of the strip casting process for given aluminum alloy. The proposal brings together two small but innovative industrial contributors focused on applied uses of machine learning and novel aluminum casting technologies, respectively. While the machine learning approach will be applied here to industrial strip-casting of aluminum, we consider this a development test-bed for procedures and processes that are directly transferable to a range of continuous production industries. To enable application across a wide range of industries, the approach developed will therefore be modular and flexible. Our goal is to address a significant knowledge gap with the potential to benefit a wide range of Canadian industries.