APPLIED ARTIFICIAL INTELLIGENCE IN FLEET CONTROL OF AUTOMATED GUIDED VEHICLES

Abstract

In recent years, Automated Guided Vehicles (AGVs) have generated much interest in the manufacturing industry, particularly in the area of automated material handling. For fleet control of AGVs, effective dispatching rules for the AGVs have to be formulated to ensure their smooth operations (i.e. picking up the parts from the source workcenters and delivering the parts to the destination workcenters) on the shop floor. Due to the complexity and dynamics involved in the fleet control of the AGVs, artificial intelligence is applied to tackle the operational problem concerning the fleet control of AGVs. Specifically, fuzzy logic is used as the basis for the fleet control of AGVs. An approach to dispatching of AGVs based on a self-adapting fuzzy method in a flexible manufacturing environment is adopted. This method allows for the possibility to formulate more versatile and flexible rules that do not specialise any more in a single criterion satisfaction. Also, the balance between rules can be precisely adapted to a production environment through real-time parameterisation from available information. The performance of this fuzzy approach to dispatching AGVs is evaluated against current approaches in the case studies, and positive results are observed. Having developed the fuzzy vehicle dispatching system, tuning methodologies are needed to fine-tune the fuzzy vehicle dispatching system. Particularly, the weights of the fuzzy rules have to be fine-tuned to improve the efficiency of the vehicle dispatching system. Genetic Algorithm (GA) and Taguchi method are separately applied to tune the weights (scaling factors) of the fuzzy rules in the fuzzy vehicle dispatching system. Then, a tuning methodology based on GA for optimal fuzzy dispatching rules is developed. The developed GA-Fuzzy vehicle dispatching system is a significantly improved version of the earlier developed fuzzy vehicle dispatching system, with higher shop floor operational efficiency. Dynamic and adaptive vehicle dispatching strategy is then implemented in the GA-Fuzzy vehicle system to fully utilise the on-line information available from the shop floor at all times. This continuously adaptive strategy is built on the fuzzy vehicle dispatching system. Simulation results on the AGVs platform show that the GA-fuzzy dispatching system, with continuous adaption capability, has outperformed other conventional approaches in the case studies. Subsequently, a new statistical approach towards vehicle dispatching for a fleet of AGVs is adopted. Using the adaptive fuzzy rules formulated earlier in the fuzzy vehicle dispatching system, the Taguchi method is applied to fine tune the fuzzy rules for optimal performance. Simulations on the AGVs platform are conducted to compare the Taguchi method to other earlier reported methods. The simulation results obtained using this hybrid Fuzzy-Taguchi method is very encouraging. Finally to explore further the potential of the Taguchi method as a search and tuning method, the Taguchi method is applied to tune the weights of a radial basis function (RBF) neural network. An alternative platform is needed to test the approach since the earlier developed AGVS (with fuzzy dispatching rules) does not use RBFs. For this reason, a small deviation is made in this thesis to consider a test platform involving high precision motion control. Specifically, the developed method (i.e. Taguchi-tuned RBF neural network) is applied to tune a composite motion controller incorporating RBF-based adaptive control. The effectiveness of using a Taguchi-tuned RBF network on a high precision motion control platform is demonstrated in the simulation results. Experimental results on a high precision motion control platform are also provided.