Ritesh A. Khire

Selection-Integrated Optimization (SIO) Methodology for Optimal Design of Adaptive Systems

Many engineering systems are required to operate under changing operating conditions. A special class of systems called adaptive systems has been proposed in the literature to achieve high performance under changing environments. Adaptive systems acquire this powerful feature by allowing their design configurations to change with operating conditions. In the optimization of the adaptive systems, designers are often required to select (i) adaptive and (ii) nonadaptive (or fixed) design variables of the design configuration. Generally, the selection of these variables and the optimization of adaptive systems are performed sequentially, thus being a source of suboptimality. In this paper, we propose the Selection-Integrated Optimization (SIO) methodology, which integrates the two key processes: (1) the selection of the adaptive and fixed design variables and (2) the optimization of the adaptive system, thereby eliminating a significant source of suboptimality from adaptive system optimization problems. A major challenge to integrating these two key processes is the selection ofappropriate fixed and adaptive design variables, which is discrete in nature. We propose the Variable-Segregating Mapping-Function (VSMF), which overcomes this challenge by progressively approximating the discreteness in the design variable selection process. This simple yet effective approach allows the SIO methodology to integrate the selection and optimization processes and helps avoid one significant source of suboptimality from the optimization procedure. The SIO methodology finds its applications in a variety of other engineering fields, such as product family optimization. However, in this paper, we limit the scope of our discussion to adaptive system optimization. The effectiveness of the SIO methodology is demonstrated by designing a new air-conditioning system called Active Building Envelope (ABE) system.

Comprehensive Product Platform Planning (CP3) Framework

Development of a family of products that satisfies different market niches introduces significant challenges to todays manufacturing industries—from development time to aftermarket services. A product family with a common platform paradigm offers a powerful solution to these daunting challenges. This paper presents a new approach, the Comprehensive Product Platform Planning (CP 3) framework, to design optimal product platforms. The CP 3 framework formulates a generalized mathematical model for the complex platform planning process. This model (i) is independent of the solution strategy, (ii) allows the formation of sub-families of products, (iii) allows the simultaneous identification of platform design variables and the determination of the corresponding variable values, and (iv) seeks to avoid traditional distinctions between modular and scalable product families from the optimization standpoint. The CP 3 model yields a mixed integer nonlinear programming problem, which is carefully reformulated to allow for the application of continuous optimization using a novel Platform Segregating Mapping Function (PSMF). The PSMF can be employed using any standard global optimization methodology (hence not restrictive); particle swarm optimization has been used in this paper. A preliminary cost function is developed to represent the cost of a product family as a function of the number of products manufactured and the commonality among these products. The proposed CP 3 framework is successfully implemented on a family of universal electric motors. Key observations are made regarding the sensitivity of the optimized product platform to the intended production volume.

PRODUCT FAMILY COMMONALITY SELECTION THROUGH INTERACTIVE VISUALIZATION

High dimensionality and computational complexity are curses typically associated with many product family design problems. In this paper, we investigate interactive methods that combine two traditional technologies — optimization and visualization — to create new and powerful strategies to expedite high dimensional design space exploration and product family commonality selection. In particular, three different methods are compared and contrasted: (1) exhaustive search with visualization, (2) individual product optimization with visualization, and (3) product family optimization with visualization. Among these three, the individual product optimization with visualization methods appears to be the most suitable one for engineer designers, who do not have strong optimization background. This method allows designers to “shop” for the best designs iteratively, while gaining key insight into the tradeoff between commonality and individual performance. The study is conducted in the context of designing a UTC product using an in-house, system-level simulation tool. The challenges associated with (1) design space exploration involving mixed-type design variables and infeasibility, and those associated with (2) visualizing product family design spaces during commonality selection are addressed. Our findings indicate a positive impact on the company’s current approach to product family design and commonality selection.Copyright

Developing a Non-gradient Based Mixed-Discrete Optimization Approach for Comprehensive Product Platform Planning (CP 3 )

The Comprehensive Product Platform Planning (CP ) framework presents a flexible mathematical model of the platform planning process, which allows (i) the formation of sub-families of products, and (ii) the simultaneous identification and quantification of platform/scaling design variables. The CP 3 model is founded on a generalized commonality matrix that represents the product platform plan, and yields a mixed binary-integer nonlinear programming problem. In this paper, we develop a methodology to reduce the high dimensional binary integer problem to a more tractable integer problem, where the commonality matrix is represented by a set of integer variables. Subsequently, we determine the feasible set of values for the integer variables in the case of families with 3 − 7 kinds of products. The cardinality of the feasible set is found to be orders of magnitude smaller than the total number of unique combinations of the commonality variables. In addition, we also present the development of a generalized approach to Mixed-Discrete Non-Linear Optimization (MDNLO) that can be implemented through standard non-gradient based optimization algorithms. This MDNLO technique is expected to provide a robust and computationally inexpensive optimization framework for the reduced CP 3 model. The generalized approach to MDNLO uses continuous optimization as the primary search strategy, however, evaluates the system model only at the feasible locations in the discrete variable space.

Model Validation and Error Modeling to Support Sequential Sampling

Several model validation and prediction error modeling techniques are studied and compared in this paper to help establish stopping criteria and identify critical regions in the design space in a sequential sampling framework. This study leads to the proposal of a two-phase sequential sampling and meta-modeling strategy, which is realized by the support of a multi-dimensional data visualization tool. These techniques have been successfully applied in the development and setup of a system-level parametric tool to support Heating, Ventilating, and Air Conditioning design. Maintaining the same level of accuracy, we observe a savings of 6–30 times the simulation effort needed for current practice. The benefits and drawbacks of the method are discussed, and opportunities are identified for future improvement.Copyright

An Optimization Based Methodology to Design Flexible Systems Subjected to Changing Operating Conditions

Flexible systems maintain a high performance level under changing operating conditions or design requirements. Flexible systems acquire this powerful feature by allowing critical aspects of their design con guration to change during the operating life of the product or system. In the design of such systems, designers are often required to make critical decisions regarding the exible and the non-exible aspects of the design con guration. We propose an optimization based methodology to design exible systems that allows a designer to effectively make such critical decisions. The proposed methodology judiciously generates candidate optimal design versions of the exible system. These design versions are evaluated using multiobjective techniques in terms of the level of exibility and the associated penalty. A highly exible system maintains optimal performance under changing operating conditions, but could result in increased cost and complexity of operation. The proposed methodology provides a systematic approach for incorporating designer preferences and selecting the most desirable design version — a feature absent in several recently proposed exible system design frameworks. The developments of this paper are demonstrated with the help of a exible three-bar-truss design example.Copyright

Comprehensive Product Platform Planning (CP 3 ) Framework: Presenting a Generalized Product Family Model

Development of a family of products that satisfies different sectors of the market introduces significant challenges to today’s manufacturing industries – from development time to aftermarket services. A product family with a common platform paradigm offers a powerful solution to these daunting challenges. The Comprehensive Product Platform Planning (CP 3 ) framework formulates a flexible product family model that (i) seeks to eliminate traditional boundaries between modular and scalable families, (ii) allows the formation of sub-families of products, and (iii) yield the optimal depth and number of platforms. In this paper, the CP 3 framework introduces a solution strategy that obviates common assumptions; namely (i) the identification of platform/non-platform design variables and the determination of variable values are separate processes, and (ii) the cost reduction of creating product platforms is independent of the total number of each product manufactured. A new Cost Decay Function (CDF) is developed to approximate the reduction in cost with increasing commonalities among products, for a specified capacity of production. The Mixed Integer Non-Liner Programming (MINLP) problem, presented by the CP 3 model, is solved using a novel Platform Segregating Mapping Function (PSMF). The proposed CP 3 framework is implemented on a family of universal electric motors.

One-Step Continuous Product Platform Planning: Methods and Applications

This chapter presents two methodologies, Selection-Integrated Optimization (SIO) and Comprehensive Product Platform Planning (CP3), which convert the inherently combinatorial product family optimization problem into continuous optimization problems. These conversions enable one-step product family optimization without presuming the choice of platform and scaling design variables. Such approaches also enable taking full advantage of continuous optimization methods.

Comprehensive Product Platform Planning (CP3) Using Mixed-Discrete Particle Swarm Optimization and a New Commonality Index

A product family with a common platform paradigm can increase the flexibility and responsiveness of the product-manufacturing process and help take away market share from competitors that develop one product at a time. The recently developed Comprehensive Product Platform Planning (CP3) method allows (i) the formation of sub-families of products, and (ii) the simultaneous identification and quantification of platform/scaling design variables. The CP3 model is founded on a generalized commonality matrix representation of the product-platform-plan. In this paper, a new commonality index is developed and introduced in CP3 to simultaneously account for the degree of inter-product commonalities and for the overlap between groups of products sharing different platform variables. To maximize both the performance of the product family and the new commonality measure, we develop and apply an advanced mixed-discrete Particle Swarm Optimization (MDPSO) algorithm. In the MDPSO algorithm, the discrete variables are updated using a deterministic nearest-feasible-vertex criterion after each iteration of the conventional PSO. Such an approach is expected to avoid the undesirable discrepancy in the rate of evolution of discrete and continuous variables. To prevent a premature stagnation of solutions (likely in conventional PSO), while solving the high dimensional MINLP problem presented by CP3, we introduce a new adaptive diversity-preservation technique. This technique first characterizes the population diversity and then applies a stochastic update of the discrete variables based on the estimated diversity measure. The potential of the new CP3 optimization methodology is illustrated through its application to design a family of universal electric motors. The optimized platform plans provide helpful insights into the importance of accounting for the overlap between different product platforms, when quantifying the effective commonality in the product family.Copyright

Economic Viability Assessment of Active Building Envelope Systems

Active Building Envelope (ABE) systems represent a new thermal control technology that actively uses solar energy to compensate for passive heat losses or gains in buildings or other enclosures. As a result, these systems are expected to eliminate the need to supply electricity to operate conventional air-conditioning systems and/or non-renewable energy sources to thermally condition buildings or other enclosures. This paper presents new findings pertinent to the development of ABE systems. Specifically, in this paper, we investigate the economic viability of ABE systems. A preliminary cost model of the ABE system is developed that combines individual cost models of its components. Different configurations of ABE systems, each comprising different heat absorbing component, are examined. The configuration that requires the least power unfortunately also results in the highest cost. A multi-objective optimization tool is used to resolve the trade-off between the power and the cost of ABE systems. Based on these optimization results, direction for future work is suggested. This paper represents an important step forward in obtaining a cost-effective ABE system that will provide an attractive alternative to current approaches.