Modern industries operate under high cost pressure coupled with ever increasing demands on their processes and products. Production processes are e.g. increasingly complex while the permitted tolerances, batch sizes and time-to-market times decrease. These partly contradictory trends require sophisticated production processes with advanced strategies for quality assurance and process control. The aim of this work is to analyse such a complex, multi-stage production process, the production of turbine blades, in terms of quality, process adjustment for small batch sizes and cost.
In the considered process, turbine blades are manufactured by forging and are cooled down in calm air to ambient temperature for subsequent machining. This significantly impedes quality control during the process due to the prevailing elevated temperature of workpieces and the consequential need for several hours of cooling before measurements can be performed. Due to small batch sizes, forging of one batch is completed within hours, possibly before quality control at the first produced workpiece takes place. This results in late verification of tolerances when all workpieces are already produced and potentially violate their tolerance limits. After forging, the focus is on verification of dimensional forging tolerances. These asymmetric tolerances allow only for additional material that is to be subsequently removed by machining. Equivalent asymmetry is encountered in the incurred cost, positive deviations increase machining cost while negative deviations cause high cost due to classification as defective.
Analysis of the production process indicated substantial process optimisation opportunities by quality control during the process. However, not only measurements at elevated workpiece temperature have to be performed, also the cooling influence on the workpiece must be predicted to make early conformance statements. This is especially crucial for the thin freeform aerofoil of turbine blades that is subject to complex geometrical distortions during cooling. Additionally, if the process parameters shall be adjusted according to measurement results, appropriate methods to account for asymmetric tolerances and cost are necessary. Adjusting process parameters during the ramp-up of a batch is necessary to setup the process for the specific product. Such adjustments slow down the production, can be costly and may require a considerable period of the production time, especially for small batches. Therefore, a method shall be developed to determine when to stop initial adjustments.
In this work, a multisensor light sectioning coordinate measuring system for dimensional measurements at elevated temperature is presented and discussed. For visualisation and measurand evaluation, an existing heuristic surface reconstruction method is adapted for enhanced surface quality on partly concave freeform workpieces as turbine blades. Its low time complexity enables realtime visualisation during measuring, allowing operators to monitor and qualitatively verify measurement results quickly. Main uncertainty contributors on the system are identified, quantified and, where necessary, corrected. In particular for freeform workpieces, the requirement for improved sensor adjustment is demonstrated. A novel method for sensor adjustment and multisensor registration is proposed, yielding a five times improvement in experiments compared to manual methods.
By the discussed corrections, process adjustment for small batch turbine blade manufacturing becomes feasible. A method to obtain the optimal number of adjustments is available from literature for a specific combination of symmetric cost model and process variation by analytic evaluation of expected cost. A novel formulation and appropriate numerical methods are proposed to evaluate expected cost with arbitrary, possibly asymmetric, cost models and process variation models. Based on this formulation, two generalised criteria when to stop adjustments optimally are presented, each exhibiting distinct advantages for specific application cases. Their performance is compared to a state-of-the-art deadband model for process adjustment, yielding down to 90% lower cost for the evaluated cases if measurements are performed during the adjustment phase only. Eventually, a novel comprehensive framework for process adjustment, incorporating the proposed methods, is discussed.