Simulation-based optimisation often considers computationally expensive problems. Successfully optimising such large scale and complex problems within a practical time frame is a challenging task. Optimisation techniques to fulfil this need to be developed. A technique to address this involves decomposing the considered problem into smaller subproblems. These subproblems are then optimised separately. In this paper, an efficient algorithm for simulation-based optimisation is proposed. The proposed algorithm extends the cooperative coevolutionary algorithm, which optimises subproblems separately. To optimise the subproblems, the proposed algorithm enables using a deterministic algorithm, next to stochastic genetic algorithms, getting the flexibility of using either type. It also includes a constructive heuristic that creates good initial feasible solutions to reduce the number of fitness calculations. The extension enables solving complex, computationally expensive problems efficiently. The proposed algorithm has been applied on automated sheet metal press lines from the automotive industry. This is a highly complex optimisation problem due to its non-linearity and high dimensionality. The optimisation problem is to find control parameters that maximises the line’s production rate. These control parameters determine velocities, time constants, and cam values for critical interactions between components. A simulation model is used for the fitness calculation during the optimisation. The results show that the proposed algorithm manages to solve the press line optimisation problem efficiently. This is a step forward in press line optimisation since this is to the authors’ knowledge the first time a press line has been optimised efficiently in this way.

This paper presents successful feed forward control of additive manufacturing of fully dense metallic components. The study is a refinement of former control solutions of the process, providing more robust and industrially acceptable measurement techniques. The system uses a solid state laser that melts metal wire, which in turn is deposited and solidified to build the desired solid feature on a substrate. The process is inherently subjected to disturbances that might hinder consecutive layers to be deposited appropriately. The control action is a modified wire feed rate depending on the surface of the deposited former layer, in this case measured as a resistance. The resistance of the wire stick-out and the weld pool has shown to give an accurate measure of the process stability, and a solution is proposed on how to measure it. By controlling the wire feed rate based on the resistance measure, the next layer surface can be made more even. A second order iterative learning control algorithm is used for determining the wire feed rate, and the solution is implemented and validated in an industrial setting for building a single bead wall in titanium alloy. A comparison is made between a controlled and an uncontrolled situation when a relevant disturbance is introduced throughout all layers. The controller proves to successfully mitigate these disturbances and maintain stable deposition while the uncontrolled deposition fails.

The accuracy of a pose measurement system, called PosEye, is investigated. PosEye is a system for real time measurement of the position and orientation, the pose, of a camera (sensor) using the information in its image. This sensor is aimed to be mounted on an industrial robot for welding. The investigation was done by comparing the PosEye system position output to that of a coordinate measuring machine. Sources of errors are identified, and suggestions for improvements are made

This paper presents novel strategies for better calibration and pose calculations of a system for determining the pose, i.e. position and orientation, of a camera. The system in question has a camera aimed to be placed on the hand of an industrial robot for welding, but is useful for any application with a need for measuring position and/or orientation. To calculate the pose of the camera circular reference points that can be recognized in the images are distributed in the working area. From their 2D image coordinates the 6D pose of the camera can be calculated. First the system is calibrated, i.e. the positions of the reference points and the camera parameters are determined. This is done by first taking images of the reference points from different locations, and then do a "total calibration" procedure to calculate the unknown parameters. For a specific system, called PosEye, it was concluded that the accuracy needs to be improved for welding applications. Also a method for making the calculations converge more easily, was needed. To meet these demands a new camera model is proposed, and three preprocessing calculation steps are presented. The new camera model increases accuracy by considering more distortion effects. The preprocessing steps give better initial values for more robust convergence of the algorithms and increased accuracy

A position and orientation (pose) measurement system is being developed. The system, called PosEye, is based on a camera and by using the information in the image, the pose of the camera taking the image can be calculated. The system is aimed to be placed on an industrial robot for welding, but it is flexible and can also be used in many other applications. The accuracy has been measured, and it is concluded that the accuracy needs to be improved for welding applications. To make the pose measurement, reference points, that can be recognized in the image, are distributed in the working area. The positions of the reference points and the parameters in a camera model are initially computed automatically from sample images from a number of directions to the reference points. After this calibration, the pose can be calculated at each sample image. For high accuracy there is a need to have a camera model that takes into account a number of distortion effects, which are further developed in this paper. The new model is used to express an optimization cost function that can be used for both the pose calculation, and the extensive calibration, that determines camera parameters in the camera model and the positions of the reference points

This paper describes a promising approach where simulations were used in the design of real-time control for automated welding. A finite element method has been used for thermal modelling of gas tungsten arc welding on a simplified test object. Measurement data for model calibration and validation was acquired through thermal imaging during weld experiments on test objects of the alloy Fe-316. An optimisation scheme for inverse modelling was employed in the calibration of the distributed weld process model. Frequency weighted model reduction and parametric system identification were applied and evaluated to get a low order model of the single-input single-output dynamics between a simulated weld heat source (actuator) and a sensor. This low order model was then successfully used for controller design where the control signal was weld current and the measured output was a moving spot temperature. Finally, the closed-loop performance was evaluated by simulation of the weld process model showing improved temperature stability relative to open loop.

Optical three dimensional scanning for weld distortion measurements have been performed for validation, inspection, general visualization and documentation of a robotized welding process. The planning, preparing and processing of the weld is done in a simulation-based concept where computer aided robotics software simulations are integrated with finite element analysis simulations with the objective to reduce global geometrical deformation during welding. The off-line programmed robot paths were used as an input for finite element calculations of temperature fields and distortion in the work piece. In order to validate the finite element model 3D-scannings have been performed before and after every single welding sequence. This paper describes a validation experiment with non-contact measurements of weld distortion and discusses limitations in optical 3D-scanning techniques used for this purpose

In this contribution, it is shown how multivariable sampled-data controllers can be generated for a set of linearized models. The size of this set makes an automatic generation of controllers more practical and less time consuming. The control strategy is robust loopshaping according to Glover/McFarlane. The generated controllers are to be used for the military turbofan engine F404-RM12 which is currently used in the Swedish air fighter JAS39 Gripen.