Pipes structures are widely spread and used in the industrial sectors, mainly oil, gas, chemical, etc. These pipes travel for hundreds of kilometres under the earth's surface; they become inaccessible and difficult to inspect at every location. Also, it isn't easy to inspect the outer surface of the pipe as they are insulated from the outer surface.
Structural Health Monitoring (SHM) using the Ultrasonic Guided Waves (UGW) method is adapted to monitor the structure. UGW testing is used to locate the presence of damage and localise the damage.
The project aims to develop a damage inspection profilometry system for a hexagonal subassembly component of a nuclear fast breeder reactor core using a structured light source/and an imaging sensor that operates through a radiation shield glass medium 80-120 cm thick, placed between the component and the source/imaging system.
The project involves the development, standardization, and implementation of the methodology for remote metrology of a hexagonal wrapper (of 500mm length) positioned at about 2.0-2.5m distance from the laser/camera system and through the shielding glass window. Expected precision 50 microns.
Diesel oil engine inspection should be achieved without stripping down the system to save inspection time. The fuel injector comes with two different configurations: the injector is parallel to the bore axis, or the injector is inclined at a specified angle to the bore cylinder axis.
The 360˚ angular extent of the bore is inspected by a borescope camera aligned axially with the bore but fitted with a mirror of 45˚ to obtain a view of the lateral surface of the bore. The images/video captured from the camera are mosaicked and displayed on the screen as an opened cylinder.
The machine is used to visually count stacked in the range of 200-micron thickness sheets with high accuracy (<0.1%). In our case, they are thin tin-coated mild steel sheets with optically challenging specular characteristics. The sheets are further not expected to be perfectly aligned. Manually counting such sheets is dangerous and time-consuming.
A naive way of counting stacked substrate would be to simply divide the overall thickness/weight by the nominal thickness/weight of one sheet in the stack. However, this method fails to achieve the required accuracy because individual sheets deviate in their thickness/weight from the nominal values because of manufacturing variations, and this could lead to counting errors greater than 1%. The machine can count a 50 cm stack in under 3 minutes. The system includes image acquisition, optical distortion correction, customized stitching, and occluded sheet compensation. The counting result is displayed on an included numeric display. (Hardware Development)
The experiment was conducted by keeping a constant compression ratio and with varying speeds; significant improvement in brake thermal efficiency and brake-specific fuel consumption was observed when compared with diesel oil.
NOx, CO, and CO2 emissions for ethanol blends are less than the diesel oil at higher loads.
Vinoothan Kaliveer, Prajwal Raphael Sequeira, Rayan Veneeth D’sa, Simmons Antony Dsilva, Sandeep B., Rolvin S. D’silva, Investigation on the Performance and Emissions Characteristics of CI Engine Using Different Blends of Waste Cooking Oil Methyl Ester-Ethanol-Diesel Oil , Energy and Power, Vol. 6 No. 1A, 2016, pp. 28-32. doi: 10.5923/c.ep.201601.05
The system needs to inspect the soft drink PET bottles and automatically discard the faulty bottle from the conveyor. The bottle sizes vary from 250ml to 1.5 litres. The faults of the bottle are under & overfilled liquid, damaged closure, and torn label.
A single high-speed camera is used to capture images on the conveyor; 400-600 bottles per minute is the conveyor travel speed. The captured images are processed; the default bottles kicked away from the conveyor using a pneumatic system.
Coded and tested gradient descent to train and validate L1 and L2 penalised linear regression from scratch.
Performed exploratory data analysis and eliminated correlated variables and imputed missing variables. I also trained and validated hyperparameters of elastic net logistic regression, neural networks, support vector machines, and random forests for a classification problem. I also checked feature importance and eliminated variables to improve model performance.
Modified the pre-existing CNN training code to classify bees versus ants to run on Google Colab to test learning rate and momentum and added weight decay for regularisation. I also pre-processed data for clustering using various transforms and trained and visualised the effect of k in k-means clustering. I also trained in principal component analysis and selected the number of dimensions based on the variance explained. I also trained kernel principal component analysis and selected the number of dimensions based on the variance explained.
Controlling the EMU robot with a hand gesture using the Inertial measurement unit sensors.
IoT-enabled presence-activated smart lighting solution for use in the Mechatronics Department of Motion Tracking (Mini Project)