Industrial Resilience: Risks and Mitigation Strategies in the Automotive Industry

这个跨学科的研究项目涉及EPSRC的几个研究领域。重点是“运筹学”，以便改进决策。通过分析生产和供应网络中的操作实践和结构模式，论文项目的目的是确定这些网络中危及产品安全的漏洞。近年来，导致许多行业高调召回的大部分操作模式尚未得到调查。最近的研究揭示了一些可能的原因，例如，越来越复杂的产品已被确定为召回率上升的来源。此外，新技术在消费品等具有关键安全要求的产品中的应用，为管理实践和工程带来了新的挑战。例如，与自动驾驶汽车技术相关的召回事件在过去10年呈指数级增长，预计还会进一步增加。许多研究学者和不同行业已经制定了应对高调召回的缓解措施的初步方法。这些措施包括政策层面的措施，例如实施新的标准和法规，以及增加可追溯性，例如通过应用RFID技术。然而，这些方法仍然是零散的，更重要的是，在汽车生产等大型工业部门，召回率上升的问题仍然很严重。我在硕士论文中对召回模式的分析表明，产品的设计缺陷和生产缺陷是产品故障的根本原因。然而，问题仍然是，从根本原因层面上，什么是有效的对策？当前的管理实践如何解决这些问题？因此，本文旨在探讨产品召回的原因，并制定提高产品安全的策略。此外，本研究将采用混合方法，研究领域“人工智能技术”将与本论文相关。人工智能分析在制造业中的大部分潜力尚未实现。最近的一项研究发现，仅在汽车和装配行业，人工智能分析每年带来的效率提升估计可达3000亿美元。在这种情况下，与产品故障相关的风险分析应用程序是多种多样的。例如，公司有时寻求反应性召回策略，而不是预防性策略，这意味着只有在安全隐患造成伤害或死亡后才宣布召回。通过预测分析等新技术，预测产品故障可以防止死亡，同时可以提高运营效率和产品一致性，在早期阶段使用信息纠正设计或生产中的缺陷。此外，今天的管理人员经常面临这样的问题：如果有产品故障的迹象，但没有具体的证据，是否应该发布召回。因此，预测算法也可以作为从业者的决策支持工具，以防潜在的故障产品进入市场。虽然完善的质量管理技术，如“失效模式和影响分析”，如今是多个工业部门部件批准的先决条件，但包括数据分析在内的新方法可能会变得更加相关。

This interdisciplinary research project concerns several research areas of the EPSRC. The focus is on "Operational Research" in order to enable improved decision-making. By analysing operational practices and structural patterns in production and supply networks, the aim of the dissertation project is to identify vulnerabilities in these networks which compromise product safety. Large parts of the operational patterns leading to high profile recalls in many industries in recent years have not been investigated. Recent research has shed some light on possible reasons, for example increasingly complex products have been identified as a source of increasing recall rates. Moreover, the application of new technologies in products with critical safety requirements, such as consumer goods, brings new challenges for management practices and engineering. For example, recalls related to autonomous vehicle technologies increased exponentially within the past 10 years and are expected to increase further. First approaches for mitigation measures to deal with high profile recalls have already been formulated by multiple research scholars and for various industries. These include measures on a policy level, such as the implementation of new standards and regulations as well as an increase in traceability, for example by applying RFID technology. However, these approaches remain fragmented and more importantly, the problem of increasing recall rates in large industry sectors such as automobile production, remain critical . The analysis of recall patterns during my master's thesis indicated that design flaws in products as well as production flaws are root causes of product malfunctions. Yet the question remains what are effective countermeasures on a root cause level? How are they addressed by current management practices? Therefore, the dissertation aims to discuss the causes of product recalls and to develop strategies for increasing product safety. Furthermore, this research will rely on a mixed method approach and the research area "Artificial Intelligence Technologies" will be of relevance for this dissertation. Much of the potential of AI analytics in manufacturing is not realized yet. A recent study finds that for automotive and assembly alone, annual efficiency gains from AI analytics are estimated to 300 $bn. In this context, analytics applications for risks related to product malfunctions are various. For example, firms sometimes seek reactive recall strategies rather than preventive strategies, meaning that recalls are announced only after a safety hazard has caused injuries or deaths. With new technologies such as predictive analytics, predicting product malfunctions could prevent deaths and could at the same time increase operational efficiency and product conformity, using information in an early stage to correct flaws in design or production. In addition, managers today are often confronted with the question whether a recall should be issued if there are indications of a product malfunction, but no specific evidence. A predictive algorithm therefore could also be a decision support tool for practitioners, in case potentially malfunctioning products entered the market. Whereas well-established quality management techniques, such as "failure modes and effect analysis" are nowadays a prerequisite for component approvals in multiple industrial sectors, new approaches that include data analytics are likely to become more relevant.