Integrating Predictive Maintenance in Food Processing Units

Food processing plants can benefit from introducing a predictive solution to limit maintenance costs and reduce downtime.

FREMONT, CA: Digital transformation occurs in the food and beverage industry. It is important for companies to collect more data about their workflow processes and use digital technologies to ensure safety and quality in food processing, packaging, and distribution. As a result of this information, companies transform their production systems and reshape the way employees, processes, and assets work in the new environment. Several technologies are already in place or are being actively developed to enable the future maintenance of plants. The list includes internet of things (IoT) sensors, machine learning analytics, digital twinning for simulations and evaluations, and alternate and virtual reality for training and instructional knowledge.

A plant's technology transformation might not be feasible for older equipment, and the new maintenance method may not be supported by older equipment. Costs are also a factor.

Preventive to predictive: The practice of preventive maintenance has become a standard among manufacturers today. Preventative maintenance involves replacing parts and components based on the calendar or usage. It is an improvement over reactive maintenance, which occurs when machines fail. Preventive maintenance eliminates approximately 80 percent of unplanned downtime. Using a preventive approach means manufacturers swap out parts before they have a chance to fail, which results in higher component costs. Additionally, it causes a decrease in production throughput because machines are taken offline for maintenance when they aren't available at optimal times.

Analysis and simulation: Detecting patterns in this vast sensor data is crucial. It is common for data trends to show dramatic changes in readings simply because of changes in machine loads. For instance, a batch mixer might experience vibration spikes near the beginning or end of a run. Consequently, engineers often set high margins in out-of-spec situations—which interferes with the ability to detect signs of trouble early on.

An experienced maintenance technician was traditionally trained to detect such signs - someone who had been working for decades. With that experience, a tech can walk through a plant, listen to machinery sounds, or collect information. From the sound and other clues, the tech could detect something wrong and take corrective action, just like someone driving a familiar car.

This method has the disadvantage of taking time to acquire experience. Another disadvantage is that it can lead to a conservative approach, such as replacing parts too early and continuing in the same way. An alternative is machine learning. With this technology, software ingests large amounts of sensor data from good and bad operations, which can be defined by creating in or out-of-spec products. Software based on this information builds models used to find problems in manufacturing daily. Machine learning has proven to be extremely useful in finding these hidden clues in manufacturing and other industries.