Study On Obtaining Four-dimensional And Full-field Vibration Information Using A Depth Camera

The machine vision technology,which features with good continuity,non-contact measurement and high dimension,is widely applied in dynamic measurement,real time monitoring,etc.It is a popular area in the mechanical engineering measurement technology communities.Acquiring rich and accurate measuring data and reducing the measuring cost as well as complexity are greatly significant for dynamic measurement.In the previous works,measuring systems based on machine vision include CCD measurement system,laser measurement system,etc.However,laser measurement system suffers high cost and operating complexity.The CCD measurement system is vulnerable to the lighting condition and greatly relies on the attaching markers that bring in limitation on measuring location.Fortunately,depth camera from Microsoft Kinect sensor based on pseudo-random infrared speckle pattern is known for its full-field acquisition of 3D information,less influence by lighting condition and low measuring cost.In this thesis,we applied the depth camera to mechanical vibrating measurement to analyze its accuracy and efficiency.Our major works are listed as follow:(1)The projection theorem of the depth camera is briefly introduced first.If suitably illuminated by an extra infrared lamp and with the IR projector blocked,a depth camera can be regarded as a normal CCD camera which only works with infrared light,and can be reliably calibrated by using the same checkerboard pattern used for the normal CCD camera calibration.(2)Principles of obtaining full-field and four-dimensional vibrating information are introduced.By fusing 3D data acquired by a depth camera and time coordinates in real time,the 4D vibration information can be recorded in the form of[X Y Z t]T.Therefore,vibration information on some certain special points or areas can be easily and rapidly extracted according to the further analysis request.(3)To improve the measuring accuracy of the depth camera,we propose a combinatorial filtering method,which means the relationship of a point to its K-neighborhood is first adjusted based on the Moving Least Squares(MLS)algorithm,and then time-domain as well as frequency-domain signal of this point itself is processed with the Fast Fourier Transformation.Comparition of amplitude-frequency analysis between the datasets before and after processing is made to validate the improvement of measuring accuracy.(4)Comparative measurements are made using the stereo vision and contact measurement to discuss the strength and weakness of applying the current depth camera to vibration measurement.Comparison includes measuring amplitude-frequency characteristic,efficiency and stability.(5)Finally,we theoretically discuss the further application of our vibrating measurement method to the diagnosis of the damage state of machine parts.Especially when the damage state is unknown,the damage locations and degree can be evaluated by the Multiple Damage Location Assurance Criterion(MDLAC)that is used to detect the changes of natural frequencies of all locations.