Research On The Process Optimization Of New Product Development Considering The Total Feedback Length Minimization

The fierce market competition requires enterprises to develop new products as faster as possible.However,the new product development processes often contain many coupled activities with complex information dependences.These activities usually follow uncertain execution orders and rework frequently,which may induce more uncertainties,project delays,cost overruns and associated risks.Hence,how to schedule coupled activities in an appropriate sequence to reduce negative impact has become an important issue for project managers.The design structure matrix(DSM)can be used to express coupled activities and information dependences.Based on DSM,optimizing developing processes with the objective of minimizing total feedback lengths of coupled activity sequences,can mitigate negative effects of feedbacks,simplify information flows and decrease unnecessary reworks,which can further reduce project duration,costs and risks.This study is dedicated to developing efficient tools for an activity sequence with minimum total feedback length so as to optimize the developing processes for product development projects.The main works and contributions of this thesis are summarized as follows:(1)The feedback length minimization problem is proved to have the optimal substructure and the criteria for sequence selection are proposed.By analyzing the structure of the mathematical model,the optimal substructure has been proven,which allows to decompose a complex problem into several smaller subproblems.As the optimal solutions of the subproblems constitute the optimal solution of the original problem,the difficulty of problem solving can significantly decline.Based on the optimal substructure,two sequence selection criteria are proposed.With these criteria,the algorithm can evaluate an activity sequence according to the quality of its subsequence.It can help the algorithm to remove low-quality activity sequences in advance and improve the efficiency of finding the optimal activity sequence.(2)The hash address-based parallel branch-and-prune algorithm for finding the activity sequence with minimum total feedback length is designed.Based on the problem property,a variety of techniques are introduced to gradually improve the algorithm design: the branch-and-prune algorithm is proposed as a prototype sorting algorithm that confirms the correctness and effectiveness of the sequence selection criteria;based on the optimal substructure,the parallel branch-and-prune algorithm that contains a parallel computing framework can construct activity sequences in forward and backward directions at the same time;multiple hash function based search approaches are tested to solve the similar nodes comparation and subsequence matching problems,and the hash address-based parallel branch-and-prune algorithm is further proposed.Experimental results indicate that this algorithm’s performance is not affected by the density of DSM and can find optimal sequences for problems up to 25 coupled activities within 1 hour,which also outperforms CPLEX and Gurobi solvers.(3)The iterative tabu search algorithm is designed for larger problems that cannot be solved by the proposed exact algorithm.Based on the iterative framework,several search and perturbation components are introduced to complete the algorithm design:three different sequence modification operators are proposed,whose neighborhood scales and the features are analyzed;a tabu search phase that employs an activity exchanging operator is designed to find the local optimal solution and strengthen the algorithm intensity;a hybrid perturbation phase is designed to improve the algorithm diversity,which selects an activity group exchanging operator and a random sorting operator in a probabilistic way to guide the search procedure into a promising neighborhood;the algorithm repeats two phases until a given stopping condition is satisfied.Experimental results indicate that the proposed algorithm can obtain highquality activity sequences within 180 seconds,and the average total feedback length of activity sequences is lower than that of algorithms in literature.(4)Several instances from real product development projects are used to verify the effectiveness of the proposed algorithms.Experimental results indicate that the hash address-based parallel branch-and-prune algorithm and the iterative tabu search algorithm can obtain high-quality activity sequences within a reasonable time,which means that both of the algorithms can be used in solving practical problems.This experiment also shows some limitations of the algorithms in practical environment,and the corresponding improvement measures are also proposed.