Predictive, data-driven product family design

Predictive design analytics is a new paradigm to enable design engineers to extract knowledge from large-scale, multidimensional, unstructured, volatile data, and transform the knowledge and its trend into design decision making. Predictive, data driven family design (PDFD) is proposed as one of the predictive design analytics methods to tackle some issues in family design. First, a number and specifications of product architectures are determined by data (not by pre-defined market segments) in order to maximize expected profit. A trade-off between price and cost in terms of the quantity and specifications of architectures helps to set the target in the enterprise level. k-means clustering is used to find architectures that minimize within architecture sum of squared errors. Second, a price prediction method as a function of product performance and deviations between performance and customer requirements is suggested with exponential smoothing based on innovations state space models. Regression coefficients are treated as customer preferences over product performance, and analyzed as a time series. Prediction intervals are proposed to show market uncertainties. Third, multiple values for common parameters in family design can be identified using the expectation maximization clustering so that multipleplatform design can be explored. Last, large-scale data can be handled by the PDFD algorithm. A data set which contains a total of 14 million instances is used in the case study. The design of a family of universal electronic motors demonstrates the proposed approach and highlights its benefits and limitations.