top of page

Predictive Maintenance

An AI vision product that is one of a kind in the Travel and Transportation industry. It aims to automate the inspection processes to increase efficiency and reduce the risk of injury during physical inspection.

Read here to see CN Rail's publishing of my product.

The problem of train maintenance in the transportation industry has always been a critical issue that directly impacts train transportation safety, efficiency, and reliability. Traditional methods of train maintenance have been reactive, meaning that maintenance activities are performed only after a fault or breakdown occurs, resulting in downtime and unexpected costs.

Problem Statement

Users

Name: James Thompson

Age: 35

 

Occupation: Train Maintenance Engineer

Education: Mechanical Engineering Diploma

Location: Montreal, Quebec

 

Goals: James is a dedicated and passionate train maintenance engineer who takes pride in his work. He is always looking for ways to improve the maintenance processes and procedures to ensure the trains run smoothly and efficiently. His main goal is to ensure that the trains are always safe for passengers to travel in.

 

Challenges:

One of the biggest challenges that James faces is dealing with unexpected issues that arise while performing maintenance on the trains. He must be able to troubleshoot problems quickly and efficiently, and come up with solutions that can be implemented in a timely manner.

 

Another challenge that James faces is the pressure to complete maintenance tasks within tight deadlines. He must prioritize tasks and work efficiently to ensure that the trains are ready for service on time.

Solution 

The AI-based predictive maintenance product utilizes advanced computer vision models to inspect the railcar fleet and predict railcar or track problems before they occur. Equipped with ultra-high-definition cameras, automated inspection portals capture a 360° view of the train as it travels through at normal track speed, enabling the product to identify and categorize defects promptly. With this comprehensive solution, the client receives detailed information on the type and severity of the defect and timely notifications to address any issues that may arise.

Benefits

Increased safety

Identify potential issues before they occur, preventative and predictive maintenance approaches can significantly reduce the risk of accidents, ensuring passenger and crew safety

Extended lifespan of equipment

Identify wear and tear on components and predict their remaining useful life, enabling maintenance teams to take corrective actions to extend their lifespan.

Reduced downtime

Enable maintenance teams to plan and schedule maintenance activities in advance, minimizing the downtime of trains for maintenance and repair.

Improved efficiency

Trains can operate at optimal levels, reducing energy consumption and increasing efficiency.

Lower maintenance costs

Reduce maintenance costs and avoid expensive emergency repairs and downtime by identifying potential issues early.

Better asset management

Provide valuable data on asset performance and maintenance needs, enabling maintenance teams to make informed decisions about asset replacement or refurbishment.

Contribution

As a product manager, I was responsible for creating the product vision, and development timeline, and overseeing a Data Scientist and Developer pod to ensure that tasks were achieved within the given timeline. Additionally, I worked closely with a team of 4 data scientists, who were responsible for researching and validating alternative ML models to be used for future use cases. To optimize the delivery of use cases, I coordinated with other pods/teams and leveraged existing models.

In representing the pod to the stakeholders, I distilled their feedback into actionable insights for the pod. This required me to acquire a comprehensive understanding of the full business and technical context of the use cases before developing the technical solution. To ensure the success of the project, I documented the technical approach and recommendations, tracking and analysis of ML model performance, potential risks & roadblocks, and characteristics of the data.

Additionally, I created the testing strategy and meta-info guidelines, which involved going through and annotating more than 10,000 images. This was an important step as it allowed me and the data scientist to identify bias and create balanced training, testing, and validation datasets that are representative of reality. Overall, my role involved managing the pod, collaborating with other teams, and ensuring the successful delivery of use cases while meeting the client's requirements.

bottom of page