Current digitalized manufacturing systems do not yet achieve the goal of smart manufacturing: precise control and agility under unexpected disruptions. Push-Pull Digital Thread is a solution concept to enable contextual data and knowledge exchange across operational and functional units in a manufacturing enterprise. The extraction of decision reasoning and functional information can be facilitated by Large Language Models processing information obtained from a decision maker at the point of decision. This concept shows a potential to address critical limitations in previous endeavours for smart manufacturing systems by building a semantically searchable and sharable knowledgebase in manufacturing systems and beyond.

Design documentation is presumed to contain massive amounts of valuable information and expert knowledge that is useful for learning from the past successes and failures. However, the current practice of documenting design in most industries does not result in big data that can support a true digital transformation of enterprise. Very little information on concepts and decisions in early product design has been digitally captured, and the access and retrieval of them via taxonomy-based knowledge management systems are very challenging because most rule-based classification and search systems cannot concurrently process heterogeneous data (text, figures, tables, references). When experts retire or leave a design unit, industry often cannot benefit from past knowledge for future product design, and is left to reinvent the wheel repeatedly. In this work, we present AI-based Natural Language Processing (NLP) models which are trained for contextually representing technical documents containing texts, figures and tables, to do a semantic search for the retrieval of relevant data across large corpora of documents. By connecting textual and non-textual data through the use of an associative database, the semantic search question-answering system we developed can provide more comprehensive answers in the context of users' questions. For the demonstration and assessment of this model, the semantic search question-answering system is applied to the Intergovernmental Panel on Climate Change (IPCC) Special Report 2019, which is more than 600 pages long and difficult to read and understand, even by most experts. Users can input custom queries relating to climate change concerns and receive evidence from the report that is contextually meaningful. We expect this method can transform current repositories of design documentation of heterogeneous data forms into structured knowledge-bases which can return relevant information efficiently as well as can evolve to embody manageable big data for the true digital transformation of design.

Increasingly shortening product life cycles, regional market challenges and unforeseeable global events require highly iterative product and production adaptions. For faster adaptation, it is necessary to have a systematic understanding of the relationships between product design and production planning. A unified model and data structure are fundamental. Basic data must be extracted from both domains and integrated for consistent product-production co-design. For this purpose, we use a biological analogy, the genome-proteome phenomenon, to model the interdependencies of product (customer needs, functional requirements, design parameters) and production (technologies capabilities, machine information, process chain alternatives). From the genome, which represents the totality of available data of product and production, we contextualize the proteome, which represents an instance of a concrete product design and the corresponding production configuration. Thereby, one gene represents one incremental information set consisting of all above mentioned product and production information for a specific product function. For each of the mentioned information domains (e.g. product requirements) within a gene, a methodology exists (e.g. NLP) to model the interlinkage to the adjacent information domain (e.g. product function). Utilizing the interdependencies and heredity of product design and production planning enables quick analysis of adaptation-induced impact which will provide enhanced competitiveness in a volatile world.

Good design practice and digital tools have enabled industry to produce valuable products. Early-stage design research involves rigorous background study of large volumes of design documentation which designers must analyze manually, to extract functional requirements which are abstracted and prioritized to guide a design. Recent advances in Machine Learning, specifically Natural Language Processing (NLP), can be applied to enhance the time-consuming and difficult practice of the human designer by performing tasks such as extracting functional requirements from long-form written documentation. This work demonstrates how extractive question-answering by neural networks can be applied to design as a tool for automating this initial step in the design process. We applied the language model BERT, fine-tuned on question-answering, to identify functional requirements in written documentation. Limitations due to wording sensitivity are discussed and an outline for training a design-specific model is discussed with a MEMS product design case. This work presents how this application of AI to design could enhance the work of human designers using the power of computing, which will open the door for learning from big data of past product designs by allowing machines to "read" them.

Nearly every artifact of the modern engineering design process is digitally recorded and stored, resulting in an overwhelming amount of raw data detailing past designs. Analyzing this design knowledge and extracting functional information from sets of digital documents is a difficult and time-consuming task for human designers. For the case of textual documentation, poorly written superfluous descriptions filled with jargon are especially challenging for junior designers with less domain expertise to read. If the task of reading documents to extract functional requirements could be automated, designers could actually benefit from the distillation of massive digital repositories of design documentation into valuable information that can inform engineering design. This paper presents a system for automating the extraction of structured functional requirements from textual design documents by applying state of the art Natural Language Processing (NLP) models. A recursive method utilizing Machine Learning-based question-answering is developed to process design texts by initially identifying the highest-level functional requirement, and subsequently extracting additional requirements contained in the text passage. The efficacy of this system is evaluated by comparing the Machine Learning-based results with a study of 75 human designers performing the same design document analysis task on technical texts from the field of Microelectromechanical Systems (MEMS). The prospect of deploying such a system on the sum of all digital engineering documents suggests a future where design failures are less likely to be repeated and past successes may be consistently used to forward innovation.

Industrial innovation has accumulated big data in the form of past design successes and failures. Designers must painstakingly identify, extract, and structure requirements from texts and drawings of archived documents to understand the past and guide future designs. This is not a trivial task for human designers, despite the digitalization of design data. This paper presents a system of “Design Reading” which takes in textual design data and applies a machine learning-based language processing model to extract a structured hierarchy of functional requirements by recursively decomposing text passages. Design Reading will benefit future design practice by learning from the past.

Measuring functional coupling in complex systems is an important task for good design practice, though historically it has been an art of subjective judgement. With the recent advancements in Deep Learning and Natural Language Processing, functional requirements (FRs) and design parameters (DPs), which are expressed as words and sentences, can be represented in a vector space. The sentence embedding model, BERT, was used in this paper to vectorize FRs and DPs, to calculate functional independence and to study how metrics for functional coupling measurement can be enhanced. It was found that semantic similarity among FRs and DPs, represented in vector space, could be used to compute quantitative values for metrics of functional independence. It was also found that design cases where coupling was unambiguous yielded the best results, while cases where laws of physics needed to define the FR-DP relationship did not transliterate well to the natural language used to express the FR-DP highlighted the limitations of the model in its current state. This study, however, demonstrates a great opportunity to develop a robust, fine-tuned design language representation model for accurately measuring functional independence as a part of our effort to enhance design intelligence.

The task of design feature transcription, or encoding the functional requirements and design parameters of a design, requires representing design data such that a machine can comprehend. Natural language processing, powered by deep neural networks trained on massive corpora of textual data, can map language into distributed vector representation space that machines can understand and retrieve. This work outlines how language models can be used to enhance early-stage design by separating the functional and physical domains, abstracting key functional requirements, and analysing systems to provide metrics for good design decision making, to facilitate a framework for hybrid intelligence.

Engineering faces many wicked problems: irreducibly interdisciplinary with multiple competing objectives, and of such large scale and complexity that will require processes to deeply rely on human insights and power of computation. The resurgence of machine learning offers the possibility for new forms of human/computer collaboration where each fuels hybrid intelligence in complementary ways. A concept of virtual design assistant (VDA) is developed as a platform to bring the hybrid intelligence in solving complex design challenges. A deep learning-based abstraction process is developed to provide VDA a function to extract structured functional requirements from fragmental design specifications and customer needs.

Vibration energy harvesters based on the resonance of the beam structure work effectively only when the operating frequency window of the beam resonance matches with the available vibration source. None of the resonating MEMS structures can operate with low frequency, low amplitude, and unpredictable ambient vibrations since the resonant frequency goes up very high as the structure gets smaller. Bistable buckled beamenergy harvester is therefore developed for lowering the operating frequency window below 100Hz for the first time at the MEMS scale.This design does not rely on the resonance of the MEMS structure but operates with the large snapping motion of the beam at very low frequencies when input energy overcomes an energy threshold. A fully functional piezoelectric MEMS energy harvester is designed, monolithically fabricated, and tested. An electromechanical lumped parameter model is developed to analyze the nonlinear dynamics and to guide the design of the nonlinear oscillator based energy harvester.Multilayer beamstructurewith residual stress induced buckling is achieved through the progressive residual stress control of the deposition processes along the fabrication steps. Surface profile of the released device shows bistable buckling of 200μm which matches well with the amount of buckling designed. Dynamic testing demonstrates the energy harvester operates with 50% bandwidth under 70Hz at 0.5g input, operating conditions that have not been demonstrated byMEMS vibration energy harvesters before.