A risk-value-based methodology for enterprise-level decision-making
MetadataShow full item record
Despite its long lasting existence, aerospace remains a non-commoditized field. To sustain their market domination, the major companies need to commit to large capital investments and constant innovation, in spite of multiple sources of risk and uncertainty, and significant chances of failure. This makes aerospace programs particularly risky. However, successful programs more than compensate the costs of disappointing ones. In order to maximize the chances of a favorable outcome, a business-driven, multi-objective, and multi-risk approach is needed to ensure success, with particular attention to financial aspects. Additionally, aerospace programs involve multiple divisions within a company. Besides vehicle design, finance, sales, and production are crucial disciplines with decision power and influence on the outcome of the program. They are also tightly coupled, and the interdependencies existing between these disciplines should be exploited to unlock as much program-level value potential as possible. An enterprise-level approach should, therefore, be used. Finally, suborbital tourism programs are well suited as a case study for this research. Indeed, they are usually small companies starting their projects from scratch. Using a full enterprise-level analysis is thus necessary, but also more easily feasible than for larger groups. These motivations lead to the formulation of the research objective: to establish a methodology that enables informed enterprise-level decision-making under uncertainty and provides higher-value compromise solutions. The research objective can be decomposed into two main directions of study. First, current approaches are usually limited to the design aspect of the program and do not provide the optimization of other disciplines. This ultimately results in a de-facto sequential optimization, where principal-agent problems arise. Instead, a holistic implementation is proposed, which will enable an integrated enterprise-level optimization. The second part of this problem deals with decision-making with multiple objectives and multiple risks. Current methods of design under uncertainty are insufficient for this problem. First, they do not provide compelling results when several metrics are targeted. Additionally, variance does not properly fit the definition of risk, as it captures both the upside and downside uncertainty. Instead, the deviation of the Conditional Value at Risk (called here downside deviation) is used as a measure of value risk. Furthermore, objectives are categorized and aggregated into risk and value scores to facilitate convergence, visualization, and decisionmaking. As suborbital vehicles are complex non-linear systems, with many infeasible concepts and computationally expensive M&S environments, a time-efficient way to estimate the downside deviation needs to be used. As such, a new uncertainty propagation structure is used that involves regression and classification neural networks, as well as a Second-Order Third-Moment (SOTM) technique to compute statistical moments. The proposed process elements are combined, and integrated into a method following a modified Integrated Product and Process Development (IPPD) approach, using five main steps: establishing value, generating alternatives, evaluating alternatives, and making decisions. A new M&S environment is implemented and involves a design framework to which several business disciplines are added. A bottom-up approach is used to study the four research questions of this dissertation. At the lowest level of the implementation, an enhanced financial analysis is evaluated. Common financial valuation methods used in aerospace have heavy limitations: all of them rely on a very arbitrary discount rate despite its critical impact on the final value of the NPV. The proposed method provides detailed analysis capabilities and helps capture more value by enabling the optimization of the company’s capital structure. A sensitivity analysis also verifies the importance of the added factors in the proposed method. The second implementation step is to time-efficiently evaluate downside deviation. As such, regression and classification neural networks are implemented to estimate the base costs of the vehicle and speed up the vehicle sizing process. Business analyses are already time-efficient and therefore maintained. These neural networks ultimately show good validation prediction root-mean-square error (RMSE), which confirms their accuracy. The SOTM method is also checked and shows a downside deviation prediction accuracy equivalent to a 750-point Monte Carlo method. From a computation time standpoint, the use of neural networks is required for a reasonable convergence time, and the SOTM used jointly with neural networks results in an optimization time below 1 hour. The proposed approach for making risk/value trade-offs in the presence of multiple risks and objectives is then tested. First, the importance of using downside deviation is demonstrated by showing the risk estimation error made when using the standard deviation rather than the actual downside deviation. Additionally, the use of risk and value scores also helps decision-making from a qualitative and quantitative point of view. Indeed, it facilitates visualization by supplying a two-dimensional Pareto frontier, while still being able to color it to observe program features and cluster patterns. Furthermore, the problem with risk and value scores provides more optimal solutions, compared to the non-aggregated case, unless very large errors in weightings are committed. Finally, the proposed method provides good capabilities for identifying, ranking, and selecting optimal concepts. The last research question presents the following interrogation: does an enterpriselevel approach help improve the optimality of the overall program, and does it result in significantly different decision-making? Two elements of the enterprise-level approach are tested: the integrated optimization, and the use of additional enterprise-level objectives. In both cases, the resulting Pareto frontiers are significantly dominating their counterparts, demonstrating the usefulness of the enterprise-level approach from a quantitative point of view. It also shows that the enterprise-level approach results in significantly different decisions, and should, therefore, be applied early in the design process. Hence, the method provided the capabilities sought in the research objective. This research resulted in contributions in the financial analysis of aerospace programs, in design under multiple sources of uncertainty with multiple objectives, and in design optimization by proposing the adoption of an enterprise-level approach.