New Latent Variable Methods for selection of raw materials, process monitoring and product quality control

如今，流程工业经常收集越来越多的数据，需要对其进行分析，以提取有用的信息，以便及时做出决策。随着人们对基于工业4.0技术的“智能工厂”概念越来越感兴趣，这一趋势将在未来持续下去。它促进了从不同设备和来源（例如智能过程传感器和执行器、分析仪器、图像等）收集数据，并以结构化的方式将它们连接起来，以提高过程性能。开发有效的方法来实现这一目标是我的研究计划的重点。我的项目的长期目标包括开发新的数据驱动和系统工程方法，以减少可变性，改善过程操作和产品质量，并将其应用于对加拿大经济重要的工业领域。其目的是帮助保持国际竞争力，加速工艺和产品开发，并更有效地利用原材料和能源。在短期内，研究将集中在进一步发展我的小组最近设计的两种新的潜在变量方法，即顺序多块PLS回归（SMB-PLS）和未消差小波变换-多元图像分析（UWT-MIA）。这些方法将适用于解决原材料选择问题，开发基于图像的传感器和工具，用于过程和质量控制目的。将开发一个通用框架，用于为原材料属性设置多变量规范，以应对工业中发现的大多数实际情况。还将提出多变量过程能力度量来比较材料供应商满足客户规格的可能性。这些工具将有助于确保顺利的过程操作，达到所需的产品质量，并减少不合规格的材料和成本。本研究还旨在开发用于食品包装或气体分离应用的聚合物薄膜的在线质量控制的先进过程分析技术（PAT）。将设计一种基于成像热成像的新型传感器，以改善薄膜的质量控制。这对于早期发现产品缺陷，快速诊断原因，并采取适当的补救措施以减少不合规格的材料和生产成本是很重要的。此外，活细胞成像工具将被提议以快速和非侵入性的方式评估细胞功能。他们将利用他们的高通量筛选平台，帮助研究人员和生物技术/生物制药行业加速开发新的细胞培养基、细胞疗法和药物。最后，在这个研究项目中培养的高素质人才将在实践中发挥领导作用，将方法和工具转移到实践中，并有助于提高加拿大工业的竞争力。

Nowadays, increasingly large amounts of data are routinely collected by process industries and need to be analyzed to extract useful information for making timely decisions. This trend will sustain in the future with the growing interest for the "smart factory" concept based on Industry 4.0 technologies. It promotes data collection from different devices and sources (e.g. smart process sensors and actuators, analytical instruments, images, etc.), and interconnecting them in a structured way in order to enhance process performance. Developing efficient methods to achieve this goal is the keystone of my research program. The long-term goal of my program consists of developing new data-driven and systems engineering approaches for reducing variability, improving process operation and product quality, and applying them in industrial areas that are important for the Canadian economy. The aim is to help maintain international competitiveness, accelerate process and product development, and make a more efficient use of raw materials and energy. On the short term, the research will concentrate on further developing two new latent variable methods recently devised in my group, namely the Sequential Multi-block PLS regression (SMB-PLS) and the Undecimated Wavelet Transform - Multivariate Image Analysis (UWT-MIA). These methods will be adapted and applied for solving raw materials selection problems, and developing image-based sensors and tools for process and quality control purposes. A generic framework for setting multivariate specifications for raw material properties to cope with most practical situations found in industry will be developed. Multivariate process capability metrics will also be proposed to compare the likeliness of suppliers of materials to meet customer's specifications. These tools will help ensure smooth process operation, achieve desired product quality, and reduce the amounts of off-specification materials and costs. This research also aims at developing advanced process analytical technologies (PAT) for on-line quality control of polymer films used for food packaging or gas separation applications. A new sensor based on imaging thermography will be devised to improve quality control of the films. This is important to enable early detection of product defects, rapid diagnosis of the causes, and taking appropriate remedial actions to reduce off-specs materials and production costs. In addition, live cell imaging tools will be proposed to assess cellular functionality in rapid and non-intrusive fashion. They will help researchers and the biotechnology/biopharmaceutical industry accelerate the development of new cell culture media, cell therapies, and medications using their high throughput screening platforms. Finally, the highly qualified personnel trained in this research program will take leadership roles in transferring the methods and tools in practice, and help increase the competitiveness of the Canadian industry.