Use this URL to cite or link to this record in EThOS:
Title: Scalable design synthesis for automotive assembly system
Author: Pal, Avishek
ISNI:       0000 0004 5371 1746
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2015
Availability of Full Text:
Access through EThOS:
Access through Institution:
Frequent product model changes have become a characteristic feature in new product development and modern manufacturing. This has triggered a number of requirements such as shortening new product development time and production ramp-up time with simultaneous reduction of avoidable engineering changes and overall vehicle development cost. One of the most significant challenges when reducing new model development lead time is the large number of engineering changes, that are triggered by failures during production ramp-up stage but are unseen during design. In order to reduce engineering changes during ramp-up stage and also increase Right-First-Time development rate, there is a critical demand for improving quality of integrated product and production system design solutions. Currently, this is obtained by carrying out design synthesis which focuses on design optimization driven by computer simulation and/or physical experimentation. The design synthesis depends on the quality of the used surrogate models, which integrate critical product variables, (also known as Key Product Characteristics (KPCs)), with key process variables (Key Control Characteristics (KCCs)). However, a major limitation of currently existing surrogate models, used in design synthesis, is that these simply approximate underlying KPC-KCC relations with any deviation between the actual and predicted KPC assumed to be a simple random error with constant variance. Such an assumption raises major challenges in obtaining accurate design solutions for a number of manufacturing processes when: (1) KPCs are deterministic and non-linearity is due to interactions between process variables (KCCs) as is frequently the case in fixture design for assembly processes with compliant parts; (2) KPC stochasticity is either independent of (homo-skedastic) or dependent on (hetero-skedastic) on process variables (KCCs) and there is lack of physics-based models to confirm these behaviour; as can be commonly observed in case of laser joining processes used for automotive sheet metal parts; and, (3) there are large number of KCCs potentially affecting a KPC and dimensionality reduction is required to identify few critical KCCs as commonly required for diagnosis and design adjustment for unwanted dimensional variations of the KPC. This thesis proposes a generic Scalable Design Synthesis framework which involves the development of novel surrogate models which can address a varying scale of the KPC-KCC interrelations as indicated in the aforementioned three challenges. The proposed Scalable Design Synthesis framework is developed through three interlinked approaches addressing each aforementioned challenge, respectively: i. Scalable surrogate model development for deterministic non-linearity of KPCs characterized by varying number of local maximas and minimas. Application: Fixture layout optimization for assembly processes with compliant parts. This is accomplished in this thesis via (1) Greedy Polynomial Kriging (GPK), a novel approach for developing Kriging-based surrogate models for deterministic KPCs focusing on maximization of predictive accuracy on unseen test samples; and, (2) Optimal Multi Response Adaptive Sampling (OMRAS) a novel method of accelerating the convergence of multiple surrogate models to desired accuracy levels using the same training sample of KCCs. GPK surrogate models are then used for fixture layout optimization for assembly with multiple sheet metal parts. ii. Scalable surrogate model development for stochasticity characterized by unknown homo-skedastic or hetero-skedastic behaviour of KPCs. Application: In-process laser joining processes monitoring and in-process joint quality evaluation. Scalable surrogate model-driven joining process parameters selection, addressing stochasticity in KPC-KCC relations, is developed. A generic surrogate modelling methodology is proposed to identify and characterize underlying homo- and hetero-skedastic behaviour in KPCs from experimental data. This is achieved by (1) identifying a Polynomial Feature Selection (PFS) driven best-fitting linear model of the KPC; (2) detection of hetero-skedasticity in the linear model; and, (3) enhancement of the linear model upon identification of hetero-skedasticity. The proposed surrogate models estimate the joining KPCs such as weld penetration, weld seam width etc. in Remote Laser Welding (RLW) and their variance as a function of KCCs such as gap between welded parts, welding speed etc. in RLW. This information is then used to identify process window in KCC design space and compute joining process acceptance rate. iii. Scalable surrogate model development for high dimensionality of KCCs. Application: Corrective action of product failures triggered by dimensional variations in KPCs. Scalable surrogate model-driven corrective action is proposed to address efficient diagnosis and design adjustment of unwanted dimensional variations in KPCs. This is realized via (1) PFS to address high dimensionality of KCCs and identify a few critical ones closely related to the KPC of interest; and (2) surrogate modelling of the KPC in terms of the few critical KCCs identified by PFS; and, (3) two-step design adjustment of KCCs which applies the surrogate models to determine optimal nominal adjustment and tolerance reallocation of the critical KCCs to minimize production of faulty dimensions. All the aforementioned methodologies are demonstrated through the use of industrial case studies. Comparison of the proposed methods with design synthesis existing for the applications discussed in this thesis, indicate that scalable surrogate models can be utilized as key enablers to conduct accurate design optimization with minimal understanding of the underlying complex KPC-KCC relations by the user. The proposed surrogate model-based Scalable Design Synthesis framework is expected to leverage and complement existing computer simulation/physical experimentation methods to develop fast and accurate solutions for integrated product and production system design.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: TS Manufactures