Presentation is loading. Please wait.

Presentation is loading. Please wait.

Deflection Compensation of a Flexible Manipulator Utilizing Machine Vision and Neural Networks Starting point flexible log lifting crane with 500 kg and.

Similar presentations


Presentation on theme: "Deflection Compensation of a Flexible Manipulator Utilizing Machine Vision and Neural Networks Starting point flexible log lifting crane with 500 kg and."— Presentation transcript:

1 Deflection Compensation of a Flexible Manipulator Utilizing Machine Vision and Neural Networks Starting point flexible log lifting crane with 500 kg and 5 m maximum lifting distance. Boom tip defined by cylinder strokes and mechanical structure

2 User interface converts global coordinates to cylinder strokes ease of positioning necessitates solving of inverse kinematics of manipulator Solving inverse kinematics link matrices and flexibility matrices produced no applicable results solution based on geometry is rigid

3 Inverse kinematic model is integrated into the control system invisible to the user fast, control running at 1 kHz Inverse kinematic

4 Position deviations due to boom flexibility joint wear imprecise transducer offset inaccurate boom geometry low amplicification of PD- controller Together with high amplification of mechanical structure max deflextion 40 mm. Deflections before compensation O is set point value X is real value obtained with camera

5 Deviation correction possible with function fitting. Problem: hard to automate Solution: Automation of function fitting possible with neural networks Problem: lack of teach data Solution: Automation of data collecting with machine vision and measuring cycle.

6 Machine vision system Sony DXC-9100P-videocamera Pentium II PC Matrox Meteor II-grabber Matrox Imaging Library + ANSI C Matlab-dSpace Interface Library (for communication with DSP-card)

7 Machine vision and measuring cycle communication protocol Positioning of boom tip stroke deviation  0.3 mm Obtain real coordinates Set point value Next step Control system Machine vision system

8 Obtaining coordinates with machine vision Red filtering of pictureForm blobs calculate weight centers Blue filtering of pictureForm blobs calculate weight centers If more than one red point -> abort Find closest blue blob to red blob Calculate picture window angle from blue blobs and transform picture to plate field coordinate system. Find closest blue blob to found blue blob Phase 1. Projecting picture to plate field coordinates Phase 2. Projecting picture to global coordinates Define camera placementFrom control system Set point value (global coordinates) Predifined data Calibration point value (global coordinates) Real value in global coordinates Note! Camera rotates between 0  -90 

9 Teach process Obtained data set point value cylinder stroke obtained real value cylinder stroke calculated from real value Feed-forward backpropagation multilayer neural network Cylinder stroke deviations at given point

10 Results Compensation Deflections after compensation Note! No decline in control speed


Download ppt "Deflection Compensation of a Flexible Manipulator Utilizing Machine Vision and Neural Networks Starting point flexible log lifting crane with 500 kg and."

Similar presentations


Ads by Google