Operating data-driven inverse design optimization for product usage personalization with an application to wheel loaders

Abstract

Traditional design requires designers to envisage a product operating environment in order to identify customer needs. Analyzing product usage context by collecting actual product operating data during the product in use empowers new opportunities for the projection of requirement specifications and understanding of use case scenarios. This paper proposes a data-driven inverse design optimization approach to provide decision support to product personalization design. A closed-loop decision-making framework is formulated by integrating forward design and inverse problem solving within a coherent framework of data-driven analysis. An application to the transmission system personalization design of wheel loaders is presented to demonstrate how personalized product usage contexts are identified through inverse analysis of product operating data under different operating conditions. A particle swarm optimization (PSO) algorithm incorporated with Simulink simulation is developed to solve the multi-objective optimization of power performance and fuel economy for wheel loaders.

Introduction

Applications that integrate Internet of Things (IoT) and industrial information integration engineering are attracting scholars' and practioners' attention [7]. The advance of smart IoT and intelligent real-time sensing technologies empowers unprecedented opportunities for analyzing engineering product usage data, mining domain knowledge, and improving design for product personalization [28]. Advanced technologies (e.g., distributed sensors, RFID) can gather processing information (e.g., system status, uncertain work requirements) accurately and in real-time, which are combined with the actual physical system to realize intelligent event-driven feedback control [6].While increasing computing power greatly expands the ability of computer-aided modeling and analysis of complex systems [10], the digitalization of intelligent sensing and smart connected products through the IoT make it possible to collect product operating data (POD) and provide real-time feedbacks to the product designers [28].

Product design is traditionally limited to demand analysis that is originated from a small set of customer survey data. The use of data analysis can provide marketers a direction to make marketing strategies [47]. Collecting large customer opinion and usage data is deemed to have great potential for identifying user behaviors, understanding customer preferences, and potential customer demands [20, 47]. For example, E-commerce platforms and social media facilitate companies’ access to the massive user generated content, which empowers a data-driven approach to continuous design improvement and next-generation product prediction [35]. By analyzing the influences of customer needs, enterprise resources, and product data in the product family implementation process; and summarizing the dynamic factors of product family evolution, Hou [12] proposed a driver analysis method of product family evolution based on chaotic dynamics.

While data-driven design makes more informed decisions possible for developing better products, enormous and multiplex user- and product-generated data brings about unprecedented challenges [23]. In addition, the loss or misinterpretation of verbally communicated requirements is a daunting problem [36]. It has been noticed that the goal of mining and analyzing customer feedback data is essentially to extract the product usage context (PUC); as data itself cannot drive design decision making, but rather it is the PUC that can generate design knowledge to facilitate the decision making process [13]. Comparing to traditional design that requires designers to envisage a product in use environment in order to identify customer needs, the POD acquisition and the PUC analysis empower new opportunities for projection of requirement specifications and use case scenarios that are closer to the real market demand [13].

Analyzing the PUC relies on comprehensive data collection from the product in use, which involves interrelationships between the product and the customers, between the product and its operating environment, and the effects of the environment on the users and on the product. The emerging product intelligence and sensing technologies make it possible to obtain real-time, continuous and accurate information of the product in its operating environment. However, different from those consumer products, complex engineering products like construction machinery cope with complex tasks and work under various operating conditions, leading to difficulties in analyzing large POD for extracting the underlying PUC, as opposite to traditional feedback data collected by running tests of an experimental prototype [45].

A complex giant system need use technologies of industrial information integration to advance and integrate the information in each relevant discipline, the applications that integrate IoT and intelligent industry can realized the the continuous improvement of intelligent manufacturing/design of products [7]. In this regard, this paper proposes a data-driven inverse design optimization approach to facilitate design knowledge discovery from POD and to provide decision support to personalization design. The purpose of inverse design is to infer the best value of model parameters in forward design from the actual operating data, rather than rely on the prior assumption of these parameters.

In addition, different from the product family and product platform design based on market research, this paper proposes a new POD mining and analysis method. Through the mining of POD, we can find the personalized use patterns under different working conditions, which can improve design by projecting the relationships between diverse PUCs and various usage personalization while optimizing the overall product performance and thus increase the performance and value of personalization.

The following section introduces the status quo of understanding customization and personalization, along with related work in relation to product usage analytics, data-driven design, and inverse optimization. Distinguishing personalization design from product customization, Section 3 formulates a closed-loop decision framework of data-driven inverse design optimization. Sections 4 introduce personalization design of the wheel loader transmission system as an example to illustrate the deployment of data-driven analysis through integrated forward and inverse design. Section 5 presents a two-stage inverse optimization model by combining a PSO algorithm with simulation, whilst results and model analysis are reported in Section 6.