Knowledge-based Methods for Integrating Carbon Footprint Prediction Techniques into New Product Designs and Engineering Changes.

Abstract

This dissertation presents research focusing on the development of knowledge-based techniques of assessing the carbon footprint during new product creation. This research aims to transform the current time-consuming, off-line and reactive approach into an integrated proactive approach that relies on using fast estimates of sustainability generated from past computations on similar products. The developed methods address multiple challenges by leveraging the latest advancements in open standards and software capabilities from machine learning and data mining to support integration and early decision-making using generic knowledge of the product development field. Life-Cycle Assessment (LCA)-based carbon footprint calculation typically starts by analyzing the product functions. However, the lack of a semantically correct formal representation of product functions is a barrier to their effective capture and reuse. We first identified the advanced semantics that must be captured to ensure appropriate usability for reasoning with product functions. We captured them into a Function Semantics Representation that relies on the Semantic Web Rule Language, a proposed Semantic Web standard, to overcome limitations posed due to the commonly used Web Ontology Language. Several products are developed as Engineering Changes (ECs) of previous products but there is not enough data to predict the carbon footprint available before their implementation. In order to use past EC knowledge to predict for this purpose, we proposed an approach to compute similarity between ECs that overcame the challenge of the hierarchical nature of product knowledge by integrating an approach inspired from research in psychology with semantics specific to product development. We embedded this into a parallelized Ant-Colony based clustering algorithm for faster retrieval of a group of similar ECs. We are not aware of approaches to predict the carbon footprint of an EC or a proposed design right after the proposal. In order to reuse carbon footprint information from past designs and engineering changes, key parameters were determined to represent lifecycle attributes. The carbon footprint is predicted through a surrogate LCA technique developed using case-based reasoning and boosted-learning.