Using data-driven maintenance to improve manufacturing processes

Unscheduled equipment downtime can be detrimental to any business. Because of this, it’s critical for machinery operators to keep both field and shop equipment running to maximize utilization and minimize costly, unplanned downtime and health, safety, and environmental risks. One way to achieve strong asset performance is through data-driven maintenance.

Traditional maintenance is built primarily on a reactive and preventive approach – when something breaks you fix it, or when it’s time to change a part, you change it. This approach can lead to downtime and lost production, safety risks, expenses related to planning, overtime, rush orders of spare parts, and inventory carrying costs, plus loss of customer satisfaction due to delays.

This post will explain predictive maintenance and the benefits of using a data-driven approach to improve asset performance in the long-term.

Predictive maintenance uses a combination of sensors, artificial intelligence, and data science to optimize equipment maintenance. It monitors data collected from assets to identify patterns that lead to potential problems or failures. With this data about your equipment, you can address issues before they happen. This ability to predict when equipment or assets need maintenance allows you to optimize equipment lifetime and minimize downtime, rather than waiting for a problem to fix or only scheduling maintenance on a fixed schedule.

Predictive maintenance can be used on equipment that you know is subject to wear-out, when replacement or servicing parts are readily available, and when the failure pattern of equipment is known. Sensors monitor the asset data, a maintenance system stores and analyzes the data, then team members act based on the data.

Examples of questions you can train your predictive maintenance solution to answer:

• What is the probability that a failure will occur within the next X hours?
• What is the remaining useful life of the asset?
• Is this asset behaving in an unusual way?
• Which asset requires servicing most urgently?

The requirements for a predictive maintenance solution vary greatly by equipment, environment, process, and organization. Below is a high-level order of what’s needed to create a solution:

1. Collect training data from your assets
2. Use the data to train a machine learning model
3. Continue collecting data
4. Input data into the machine learning model
5. Identify predicted failure cases
6. Plan and act on the insights

Data lies at the heart of your predictive maintenance solution. To detect maintenance problems, you’ll need to collect the following information from your asset:

• Machine information – engine size, make, model, etc
• Telemetry data – sensor data such as temperature, pressure, vibration, fluid properties, operating speeds
• Repair history of a machine and runtime logs
• Failure history of a machine or related components

In many cases, sensor data alone might not be enough to identify an equipment failure. At times external data might be needed to “flag” a machine’s state as a failure state, or a secondary source of information, such as an operator capturing the failure case through a different system such as an ERP. The more data you have to contribute to the model, the better your solution will be at predicting issues.

After you have the appropriate data, you need to train the machine learning model on what constitutes a success and a failure, then continuously update the model based on new data. Below is an overview of creating and training a machine learning model:

1. Ingest data
2. Create, train, and test data sets
3. Determine features
4. Select algorithms
5. Train models
6. Compare models
7. Publish model
8. Repeat

Predictive maintenance can be offered by manufacturers directly by monitoring the data coming from its own operations. However, there are other way to capitalize on predictive maintenance capabilities through new business opportunities and revenue streams. Examples include:

• Adding value for customers by offering predictive maintenance services for its products
• Offering products under a Product-as-a-Service (PaaS) model where customers “subscribe” to the product instead of purchasing it outright. Under this model, the product won’t generate revenue when the product isn’t working.
• Providing predictive maintenance products and services for products manufactured by one or more manufacturers

Once you’re ready to implement a predictive maintenance strategy at your manufacturing or other organization, what’s the best technology to use? Microsoft Azure offers a range of artificial intelligence solutions and platforms for developers and data scientists. Azure Machine Learning is a cloud service for accelerating and managing the machine learning project lifecycle. Machine learning professionals, data scientists, and engineers can use it in their day-to-day workflows for training and deploying models, and managing operations.

Rand Group is a certified Microsoft partner who can help ensure all your technology solutions are fully integrated for a seamless data flow. Our team of software engineers are trained in Azure Machine Learning and have experience implementing connected field service solutions for companies in the manufacturing, oil and gas, construction, other industries. Contact Rand Group to get started today.