Automatic fault and performance loss detection of industrial chlorine electrolysis reactors at R2

电解氯化钠水溶液（或盐水）生产氯碱是世界上最大的工业规模的电合成之一。有超过1000个独立反应器的工厂，其中0.2毫米的薄膜分离氯气和氢气，是常见的。氯碱生产过程必须得到全面控制，以避免任何可能导致爆炸、高毒性气体释放、非常昂贵的电池组件不可逆转的损坏以及巨大的维护成本和生产损失的错误操作。在R2，这是加拿大在该领域的领先企业。专家系统与公知的机器学习技术相结合，以便监控和检测任何异常的性能和操作条件。由于所获取数据的性质，这种组合系统面临一些科学挑战。因此，本项目的目标是解决这些挑战，并提出一些基于机器学习领域的最新进展的解决方案，在本研究中，我们试图研究这些挑战并提出解决方案，以便通过使用机器学习技术构建一个预警和决策支持系统。该系统将检测和预测任何异常的操作条件，并建议操作员必须采取的最佳措施。

The production of Chlor-Alkli by using electrolysis of aqueous solutions of sodium chloride (or brine) is one ofthe largest industrial scale electro-synthesis worldwide. Plants with more than 1000 individual reactors, inwhich 0.2 mm thin membranes separate chlorine and hydrogen, are common. The Chlor-Alkali manufacturingprocess must be fully controlled in order to avoid any wrong operation which can cause explosions, highlytoxic gas releases, irreversible damages of very expensive cell components and dramatic maintenance costs andproduction loss.At R2, which is a leading Canadian enterprise in this field, expert systems are combined to well-knownmachine learning techniques in order to monitor and to detect any abnormal performance and operatingconditions. This combined system faces some scientific challenges due to the nature of the acquired data. Thus,the objective of this project is to address these challenges and to propose some solutions that are based on thelatest advancement in the field of machine learning.In this research, we seek to study these challenges and to propose solutions, in order to construct a warning anddecision support system by using machine-learning techniques. This system will detect and predict anyabnormal operating conditions, and will advise the operator of the best possible action that must be taken.