Liu Haijiang

Forecast Method of Customer Needs Volatility to Personalized Product

To capture and forecast the volatility of customer needs, this paper proposes a forecast method within the framework of QFD (Quality Function Deployment), based on CTS (compositional time series) and VAR model (vector auto-regression model). The CTS formed by customer needs importance rating sampling within a period of time are treated as the basis to predict the future customer needs. Firstly, the CTS are transformed from the simplex space to the real domain. Then, the VAR model is established based on the time series obtained in the real domain. This model is used to accurately forecast beyond the sample and the predictive result is transformed back to the simplex space to obtain the predictive customer needs importance rating time series. Based on the predictive customer needs importance rating, the design attributes predictive priorities are calculated, which can guide the resources allocation in the development of personalized product, to provide better personalized product that is more in line with future customer needs. The case shows that the proposed method is effective.Copyright

Material and Structural Optimization for Engine Hood Inner Panel of Car Body Aimed to Lightweight

A method of lightweight design based on material selection and structural optimization is developed for engine hood inner panel. Introduced the decision-making mechanism of multi-objective satisfaction degree to lightweight material selection, built the model of finite element analysis and computer experiment analysis of the strength, stiffness and modal properties respectively to the different lightweight material for selected, Programmed with MatLAB and realized the lightweight material optimal selection. Then performed topology structural optimization to the inner panel with the selected lightweight material properties, which realized the integrated performance improved to different level and the weight of the inner panel reduced by 30.4%. The case study of lightweight design based on lightweight material selection and structural optimization shows the feasibility and potential of the method, and it provide an reference for the lightweight design of car body.

A New Supervised Spiking Neural Network

A more computational spiking neural network, PTSNN, was proposed. In PTSNN, the synaptic connection weights between neurons were set to one. Network runs through modulating the PSP location in timeline of each neuron by adapting their accepted time make the network spike at the right time so that meet the requirement of classification. The weight modulating of PTSNN is determined by the error of actual spike time and expectation time as thus avoid calculating the derivative of error function which is often used in other SNNs. The PTSNN has more computational advantage. We perform experiments for the classical Iris dataset problem with less neurons compare to other neuron networks and the results show that it is capable to classify data set on non-linearly problem with convergence accuracy comparable to traditional sigmoidal network and other spiking neural networks. The proposed network is promise in classification problems.

The extraction of welding type for body in white based on association rules

To extract the reusable process knowledge of body in white (BIW) from process data, the association rule is employed to capture typical welding type. An association rule model for typical welding type acquisition is established. The attributes related to welding type are classified and quantitative attributes are partitioned into several intervals. Apriori algorithm is applied to extract the frequent itemsets. The strong rules are generated according to the threshold of confidence. Finally, a computational example mining typical welding process is analyzed. The result indicates that the approach can capture typical welding type effectively.