Technology Research Of Multi-sensor Integration And Measured Data Fusion For Reverse Engineering

In the mass manufacturing of marine products and the measurement of key parts such as the turbine blades, the part shapes always need to be rapidly measured and timely evaluated, and a proper action is taken to reduce the percentage of waste product and potential safety accidents. Contact and non-contact measuring are two usual shape gauging approaches. Tactile probing is a typical contact measuring way, and very accurate and repeatable, but slow on sensing speed. Laser-based non-contact scanning is very efficient and has a high sensing speed (hundreds of thousands of points per minute), but vulnerable in shiny surface regions due to the specular reflection, in high curvature areas, or in parts of complex shapes due to occlusion, which leads to incomplete data are acquired. Though each individual sensor has its own strength and weakness, these two kinds sensing approaches have complementary characteristics. And integrate multiple sensors to efficiently complete complex free-form shapes measuring now becomes a hot research problem. In this thesis, multi-sensor extrinsic calibration, measured data fusion, multi-sensor integration based on iterative fusion and registration, and a flexibly controlled multi-sensor system are deeply discussed and details as follows.A precise sinusoidal wave part based multi-sensor calibration approach is presented. In which the measuring principle of a several of sensors is surveyed, based on a basic idea that calibration is more accurate when the calibrated shape has more variation, a precise sinusoidal wave part with two-directional variations, and multiple wave-crests and wave-troughs is used to calibrate included tactile probe, laser point-probe, and laser-line sensor by combining a rotary and three-directional moving stages, which leads to calibrated data accurately describe shapes. The measuring result is validated through a accurately manufactured plane and shows that each individual sensor is well calibrated.A new Kalman filter based mulitsensor data fusion approach is presented to statistically estimate the underlying shape, which not only combine the statistic characteristics of different sensors, but also include a-priori shape estimate, and finally implement the stable estimate of data points of multiple sensors.The incremental and batch processing mode of the Kalman filter is deduced for isotropic multi-sensor measuring data fusion, and the equivalent condition of two modes is presented , which enable the optimal selection of multisensor data fusion mode according to the initial measuring status. Then it can be used to efficiently process a single measured point from tactile sensor or a batch of data points from a laser scanning sensor.Meanwhile, the fusion formula of the Kalman filter for anisotropic statistical characteristics data points from multiple sensors is given, which enable the Kalman filter approach supports the surface estimate of more general measuring data.In order to build a unified surface description for the measured shape of part, an novel iterative fusion and registration approach is proposed. In such approach, the statistic property of the Kalman filter is employed to fuse the different characteristics data points from multisensor and obtain an accurate surface estimate. The iterative closest point registration approach is used to register the other sensing data points onto such accurate surface. The registered data withdraw from the fused surface guarantees such iterative fusion and registration process converges to a local optimal solution.A high-level script programming based approach is presented flexibly control the measuring process for an integrated multisensor system including several sensors and motion stages, which enable the measuring control of a tactile probe, a laser point sensor, a laser line sensor and an area laser sensor, and combined movement or rotation of moving stage and rotary stage.Based on the above key technologies, a multi-sensor system for complex shape measuring is built including a tactile probe, a laser point sensor, a laser line sensor and an area laser sensor, and motion states ( three moving stage (X, Y, Z directions ) and a rotary stage). In addition, a prototype software system is developed for such multisensory hardware system for multi-sensor measurement calibration, registration, data-visual and surface statistical evaluation. Finally, some examples are demonstrated to validate the hardware and the prototype software system.