No abstract

Lack of planning when changing requirements to reflect stakeholders’ expectations can lead to propagated changes that can cause project failures. Existing tools cannot provide the formal reasoning required to manage requirement change and minimize unanticipated change propagation. This research explores machine learning techniques to predict requirement change volatility (RCV) using complex network metrics based on the premise that requirement networks can be utilized to study change propagation. Three research questions (RQs) are addressed: (1) Can RCV be measured through four classes namely, multiplier, absorber, transmitter, and robust, during every instance of change? (2) Can complex network metrics be explored and computed for each requirement during every instance of change? (3) Can machine learning techniques, specifically, multilabel learning (MLL) methods be employed to predict RCV using complex network metrics? RCV in this paper quantifies volatility for change propagation, that is, how requirements behave in response to the initial change. A multiplier is a requirement that is changed by an initial change and propagates change to other requirements. An absorber is a requirement that is changed by an initial change, but does not propagate change to other requirements. A transmitter is a requirement that is not changed by an initial change, but propagates change to other requirements. A robust requirement is a requirement that is not changed by an initial change and does not propagate change to other requirements. RCV is determined using industrial data and requirement network relationships obtained from previously developed Refined Automated Requirement Change Propagation Prediction (R-ARCPP) tool. Useful complex network metrics in highest performing machine learning models are discussed along with the limitations and future directions of this research.

Prior research performed by Morkos [1], culminated in the automated requirement change propagation prediction (ARCPP) tool which utilized natural language data in requirements to predict change propagation throughout a requirements document as a result of an initiating requirement change. Whereas the prior research proved requirements can be used to predict change propagation, the purpose of this case study is to understand why. Specifically, what parts of a requirement affect its ability to predict change propagation? This is performed by addressing two key research questions: (1) Is the requirement review depth affected by the number of relators selected to relate requirements and (2) What elements of a requirement are responsible for instigating change propagation, the physical (nouns) or functional (verbs) domain? The results of this study assist in understanding whether the physical or functional domain have a greater effect on the instigation of change propagation. The results indicated that the review depth, an indicator of the performance of the ARCPP tool, is not affected by the number of relators, but rather by the ability of relators in relating the propagating relationships. Further, nouns are found to be more contributing to predicting change propagation in requirements. Therefore, the physical domain is more effective in predicting requirement change propagation than the functional domain.

Design for Assembly (DFA) time estimation method developed by G. Boothroyd and P. Dewhurst allows for estimating the assembly time of artifacts based on analysis of component features using handling and insertion tables by an assembler, who is assumed to assemble the artifact one-part-at-a-time. Using the tables, each component is assigned an assembly time which is based on the time required for the assembler to manipulate (handling time) and the time required for it to interface with the rest of the components (insertion time). Using this assembly time and the ideal assembly time (i.e. the absolute time it takes to assemble the artifact, assuming each component takes the ideal time of three seconds to handle and insert), this method allows to calculate the efficiency of a design’s assembly process. Another tool occasionally used in Design for Manufacturing (DFM) is Failure Modes and Effects Analysis (FMEA). FMEA is used to evaluate and document failure modes and their impact on system performance. Each failure mode is ranked based on its severity, occurrence, and detectability scores, and corrective actions that can be taken to control risk items. FMEA scores of components can address the manufacturing operations and how much effort should be put into each specific component. In this paper, the authors attempt to answer the following two research questions (RQs) to determine the relationships between FMEA scores and the DFA assembly time to investigate if part failure’s severity, occurrence, and detectability can be estimated if handling time and insertion time are known. RQ (1): Can DFA metrics (handling time and insertion time) be utilized to estimate Failure Mode and Effects scores (severity, occurrence, and detectability)? RQ (2): How does each response metric relate to predictor metrics (positive, negative, or no relationship)? This is accomplished by performing Boothroyd and Dewhurst’s DFA time estimation and FMEA on select set of simple products. Since DFA metrics are based on combination of designer’s subjectivity and part’s geometric specifications and FMEA scores are based only on designer’s subjectivity, this paper attempts to estimate part failure severity, occurrence, and detectability less subjectively by using the handling time and insertion time. This will also allow for earlier and faster acquisition of potential part failure information for use in design and manufacturing processes.

Requirements play very important role in the design process as they specify how stakeholder expectations will be satisfied. Requirements are frequently revised, due to iterative nature of the design process. These changes, if not properly managed, may result in financial and time losses leading to project failure due to possible undesired propagating effect. Current modeling methods for managing requirements do not offer formal reasoning necessary to manage the requirement change and its propagation. Predictive models to assist designers in making well informed decisions prior to change implementation do not exist. Based on the premise that requirement networks can be utilized to study change propagation, this research will allow for investigation of complex network metrics for predicting change throughout the design process. Requirement change prediction ability during the design process may lead to valuable knowledge in designing artifacts more efficiently by minimizing unanticipated changes due to mismanaged requirements. Two research questions (RQs) described are addressed in this paper: RQ 1: Can complex network centrality metrics of a requirement network be utilized to predict requirement change propagation? RQ 2: How does complex network centrality metrics approach perform in comparison to the previously developed Automated Requirement Change Propagation Prediction (ARCPP) tool? Applying the notion of interference, requirement nodes in which change occurs are virtually removed from the network to simulate a change scenario and the changes in values of select metrics of all other nodes are observed. Based on the amount of metric value changes the remaining nodes experience, propagated requirement nodes are predicted. Counting betweenness centrality, left eigenvector centrality, and authority centrality serve as top performing metrics and their performances are comparative to ARCPP tool.