Explainable AI system to rationalise accelerated decision making on automotive component performance and manufacturability

计算机辅助工程(CAE)极大地改变了工业产品、部件和系统的生产方式，降低了开发成本、风险和上市时间，同时提高了产品质量和设计精度。研发通过金属冲压和压铸制造的汽车零部件是汽车行业最常见的制造方法，代表着全球65亿美元的产业。然而，模拟塑性应变、厚度、硬度分布和其他关键特性对零部件可制造性的情景分析需要数百万次模拟和大量熟练的工程人员。随着向零排放汽车过渡的步伐加快，英国每年仍缺少37-5.9万名3级以上工程师(英国工程现状-2019)。人工智能可以彻底改变CAE。有了它，工业界可以以数千倍的速度运行复杂的设计方案，为最复杂、最具附加值的任务腾出宝贵的工程技能。Monolith的专有人工智能引擎是在几家领先的汽车行业领先者的参与下开发的，让工程师可以基于重复性任务和历史数据构建专家模拟。然而，与其他人工智能一样，它也受到黑匣子困境的阻碍，即工程师缺乏对人工智能输出做出反应的不受限制的自由，因为他们无法合理化/控制潜在的假设。无法解释人工智能的基本假设和新组件设计的可制造性因素尤其阻碍了在严格的经济和/或监管限制下就组件可行性做出关键任务的通过/不通过决策。该项目建立在尖端学术理论的基础上，以世界上第一个考虑可制造性的复杂3D设计的人工智能可解释性工具为基础。Monolith将与伦敦帝国理工学院(ICL)的世界级可制造性专家合作，开发一种快速预测组件可制造性和性能的系统，并就人工智能如何得出结论向工程师提供明确的反馈。该平台将在该项目中与一家行业领先的制造商进行评估。到2027年，这项技术有可能产生1990万GB的累计利润(70%的出口)。在一个受到严格监管和审查的行业中，它将有助于实现人工智能的全部潜力，并极大地加快工业产品的开发。在包括航空航天和能源在内的多个行业，都有很强的翻译潜力。

Computer-aided-engineering (CAE) has dramatically changed how industrial products, components and systems are produced, reducing development costs, risk, and time to market, whilst improving product quality and design accuracy. R&D into auto-components manufactured by metal stamping and die casting, the most common manufacturing methods in the automotive industry, represents a $6.5Bn global industry. However, scenario analysis simulating plastic strain, thickness, distribution of hardness and other key properties on component manufacturability requires millions of simulations and much skilled engineering labour. With the transition to zero emission vehicles fast gathering pace, the UK still suffers from an annual shortfall of 37-59K level 3+ engineers (UK-State-of-Engineering-2019).AI can revolutionise CAE. With it, industry can run complex design scenarios many thousands of times faster, freeing up precious engineering skills for the most complex, value-added tasks. Monolith's proprietary AI engine, developed with input from several leading automotive industry leaders, lets engineers build expert simulations based on repetitive tasks and historic data. However, like other AI, it is hampered by the 'black box' dilemma, whereby engineers lack uninhibited freedom to respond to AI outputs because they cannot rationalise/control underlying assumptions. Inability to explain AI's underlying assumptions and factor in manufacturability of new component designs particularly hinder mission-critical go/no-go decisions on component feasibility made under stringent economic and/or regulatory constraints.The project builds on cutting-edge academic theory to craft the world's first AI-explainability tool for complex 3D designs with manufacturability taken into account. Working with world-class manufacturability experts at Imperial College London (ICL), Monolith will develop a system that rapidly predicts component manufacturability as well as performance, and that provides clear feedback to engineers on how the AI arrived at its conclusions. The platform will be evaluated with an industry-leading manufacturer in the project. The technology has potential to generate £19.9M cumulative profits (70% exports) by 2027\. In a heavily regulated and scrutinised industry, it will help realise AI's full potential and greatly accelerate industrial product development. There is strong translational potential in multiple sectors including aerospace and energy.