Reducing Equipment Downtime and Maintenance Costs with an AWS-Based Predictive Maintenance Platform

DataArt’s Solution

DataArt has designed, developed, and deployed a predictive maintenance platform for this customer to reduce downtime and maintenance costs and maximize the operational effectiveness of the client’s conveyor systems. The DataArt team consisted of a solution architect/machine learning expert, a data analyst, a back-end developer, a front-end developer, and four IoT developers.

The cloud-based predictive maintenance platform is capable of forecasting component degradation, so that the needed maintenance can be performed prior to equipment failure. We chose to deploy Amazon Web Services (AWS), a leading provider of the world’s most diverse and advanced cloud services. As an AWS Consulting Partner, DataArt has a vast experience in designing and implementing applications in the cloud while following modern best practices and using AWS’s purpose-built tools.

The predictive maintenance platform we created for our client was designed and implemented as a collection of microservices. Our goal was to deliver a highly extensible architecture, while serverless first was adopted as the main architectural approach.

Each predictive model is based on the unique characteristics of every component part of the conveyor system, including its surrounding conditions. The predictive maintenance system for one conveyor is comprised of more than 220 conveyer/facility-specific machine learning models. Even an identical conveyor set up in a different facility requires different models to account for different conditions in the external environment. Both the client and the facility’s maintenance staff interact with the system through a web-based dashboard.

This large-scale project, supporting four fully operational facilities in North America, was completed in one year.

While developing this solution, DataArt had to meet several challenges:

1. Connectivity. During planning, the DataArt team found that conveyor systems were being installed in facilities with little or no connectivity and limited opportunity for wired connections. We introduced a wireless mobile connectivity solution to route the data via gateways to the cloud, and then to cloud storage. To ensure reliability in the event of gateway failure, the team built mechanisms for gateway switching.
2. Expending supported hardware. Another challenge was that the team had to go beyond its usual scope of software work and tackle hardware development. To do this, our team installed sensors on all of our client’s degradable conveyor parts, including motors, gearboxes, and bearings. That allowed us to measure variables like temperature, vibration, conveyor speed, power consumption, air flow, pressure, and other important variables.
3. Creating the right ML flow. In addition to capturing real-time sensor data, the team gathered historic data like hardware-specific demographic information, weather and geographical data, inspection results, technical manuals, and maintenance reports.
4. Creating a model for each part in changing the environment. DataArt identified the best algorithms for training machine learning models to predict potential failures of each component part with the highest possible accuracy. Several machine learning models were developed for each piece of the conveyor system. Every model was uniquely responsible for monitoring a particular variable (temperature, vibration, power consumption, etc.). Our models were also trained to identify the range of “normal” behavior in a conveyor system, which would allow our client to determine whether any variances from the normal range are significant enough to warrant alerts, and determine their appropriate level of urgency.

AWS Services

The AWS services utilized are:

Solution Advantages

DataArt’s serverless, cloud-based solution combines smart sensor technology, advanced data analytics, AI models, and machine learning algorithms. Our system allows the client’s customers to protect their high-value assets from unplanned downtime, maximizing equipment utilization, operational efficiency, and overall equipment effectiveness (OEE). The predictive maintenance system provides continuous monitoring of critical equipment, a real-time view of facility asset health, tools for data storage, analysis, and visualization, and a powerful alerting feature.

Alerts for units at risk of failing are generated in real time. This enables plant maintenance staff to address equipment issues before a failure occurs, which prevents revenue loss.

The solution has the following characteristics:

• Serverless as the key architectural approach
• 99.99 high availability AWS cloud-scale data store
• Real-time event data streaming
• Cost-effectiveness, no fixed costs, and low variable costs
• Automatic disaster recovery

Business Outcomes

Thanks to our system’s remote equipment monitoring and automatic alerts, our client needs to commission repairs when needed. This reduces maintenance costs and allows for the most efficient allocation of maintenance resources.

Predictive maintenance also strengthens the client’s brand, it differentiates the systems and services from our client’s competitors, very few of whom offer predictive maintenance for conveyors.

The client can monetize the predictive maintenance solution by offering their customers additional analytics-driven services, including multiple dashboards, optimized maintenance schedules, or a technician dispatch service.

Some of the benefits of the predictive maintenance system include:

• 20% reduction in spare part inventory
• 16% reduction in overtime expenses
• 20-40% increase in machine life
• 5-10% increase in overall productivity