Learning and detecting defects in industrial sprays- ON-460Desired discipline(s): Engineering - computer / electrical, Engineering, Engineering - mechanical, Computer science, Mathematical Sciences
Company: Mazlite Inc.
Project Length: 4 to 6 months
Preferred start date: 05/10/2021
Language requirement: English
Location(s): ON, Canada
No. of positions: 1
About the company:
Mazlite has developed an IoT sensor for defect identification and process optimization of industrial sprays. Historically, spray analysis technology was only available for R&D facilities, but our mission is to bring this technology to the assembly line. Our major applications are spray painting in automotive manufacturing and spray drying in pharmaceutical manufacturing.
Our sensor is an imaging device and directly captures images of the spray which are then processed in real-time to detect issues, flag potential errors, and optimize the spray process. Mazlite is both a software and a hardware company. We have developed cutting-edge computer vision software, firmware to control our hardware, and a unique IoT platform to operate the devices and manage data. We also design and build our custom sensors which includes industrial cameras, optics, lasers, and a sophisticated mechanical housing so the sensor can operate in hazardous locations.
Please describe the project.:
Not all spray defects can be detected by current computer vision algorithms. There are certain features of a spray that people who are spray experts can observe and manually flag as being issues but our algorithms are not able to detect due to the complexity of the images. Having software that would automatically detect these features would enable our product to identify a much wider range of defects, and potentially reduce the time and amount of data needed to make a diagnosis (in a high paced manufacturing environment, reducing the time and data needed to detect an issue is extremely important).
The main goal of the company is to develop an AI model that is able to take image data and detect if there are issues with the spray. The end model will likely be a binary classifier that will sort the images into “good” or “bad” categories with reasonable accuracy. The work involved in this project will be to determine a suitable model, alter the model for our purposes, clean and organize data, train the model, assess accuracy, and a variety of other tasks related to the work as needed.
The methodology and techniques to be used will be the decision of the researcher. The researcher will work closely with our scientists and product managers to determine what the final goal will be, and monitor the project to see if adjustments need to be made as the project progresses.
Depending on the success of this project, there will be opportunities for further projects to extend the previous work or to build additional
models that target different types of defects.
Mazlite is looking for a PhD level candidate with experience in developing machine learning models from the initial conception stage to having a fully trained and working model. We will also consider a masters student with significant and relevant industry experience.
The candidate should be independent as they will be the expert in machine learning and will be responsible for the technical development strategy. The candidate should also be organized as they will be required to provide weekly or bi-weekly progress updates to present progress and metrics. Experience in model deployment is not a requirement, but would be an asset.
There are no hard software skill requirements – clearly a PhD level candidate in machine learning will have the necessary toolkit. We largely work in Python and store data on AWS, but are considering Hadoop and other technologies. However, we are flexible and will adopt whatever the right stack or tools to use are.