Predictive maintenance: Delivering impactful business outcomes through Data & Analytics (Part one)

The growth of smart technologies has opened the door to what is commonly referred as Industry 4.0, encompassing factories that use machines augmented with wireless sensors enabling the connected systems’ monitoring and automated decision-making processes throughout the production lines. The technology behind wireless sensors and its interfaces is known as the Internet of Things (IoT), where large amounts of data is collected from connected devices to monitor equipment, develop analytics platforms and take advantage of Artificial Intelligence and Machine Learning models to improve production efficiency.

A specific application of connected devices is predictive maintenance. This technology is a piece of the Industry 4.0 puzzle that allows to predict when a machine is likely to fail, for manufacturers this is directly related to operational efficiency.

IoT and Connected Systems

The foundation of smart factories starts with IoT: A series of connected devices (sensors, actuators, computing devices) that can be integrated in any mechanical or digital machine and render the ability to autonomously (without human interference) exchange data between a network of machines. In a standard configuration it is imperative that each of these devices have a unique ID so that all the activity within an IoT network can be monitored and manipulated.

In the manufacturing context, connected systems start by having a series of devices wirelessly connected. These devices can be sensors (measurement characteristics), actuators (perform an action based on an input), hybrid, databases or computers/computing devices. They have the ability to measure values stream in real time and collect data related to the operational performance of the machine – temperature, pressure, potential, humidity, other – which is posteriorly stored in a database. This data can be processed to determine the optimal operational configuration, device condition (refer to Predictive Maintenance) and automated control.

Data processing is normally carried out by a cloud-based engine (such as AWS, Google Cloud or Microsoft Azure) or on-premise computing unit, that manages all the interactions in the device network. As an example: it is known that above a certain temperature threshold a machine will have a high risk of malfunction. When an irregular temperature value is streamed through the computation unit it will send a signal to stop the process on that machine and all the others that depend on it, in order to ensure a safe production break.

A simple configuration of a connected systems’ setup at a manufacturing plant can automate several conditions known by the plant managers or process engineers, given a machine-machine interaction map. By applying machine learning algorithms on the collected data, computers are able to find more complex relationships within the connected machines’ network and improve the operational efficiency of a factory. In summary, if device behavior and production line outputs (percentage of defects, number of breakdowns, etc.) are measurable, statistical tools have a direct impact on improving manufacturing.

From simple factory settings to more complex and modular structured factories, it is possible to monitor systems, make decentralized decisions and establish a successful communication between machines and human operators to improve several processes across the manufacturing to supply value chain.

Predictive Maintenance

Predictive maintenance scope is to predict the condition of a piece of equipment in order to determine when maintenance should be performed. Since this approach targets maintenance based on the condition of the equipment, its added value is related to cost-savings when compared to time-based scheduled maintenance or preventive maintenance. By avoiding unnecessary maintenance this approach results in increased factory uptime and longer equipment lifespan, since its normally able to target defects or small repairs based on sensor measurements.

A predictive maintenance setup requires run-to-failure data to enable statistical analysis of condition and operational patterns. The more data available the more accurate the model is expected to be.

Analyzing data relative to equipment wear and failure enables the detection of anomalies, optimized scheduling and recovery plans, resulting in:

Minimized downtime

Less operational costs

Extended equipment lifetime

Safer environment

Increased product quality

The predictive maintenance monitoring suite setup can be divided in 3 parts:

Data Acquisition

Ensuring right quality of collected data is the most important in any data application. Regardless of the high specifications of an analytics platform, poor data translates into inaccurate or wrong results. To validate the data acquisition process, a clearly defined data model and cleansing process is required that always ensures formatted and congruent data.

Predictive Model

A machine learning algorithm is used to predict the machine condition based on a set of run-to-failure data points. Through data streaming technologies (Apache Kafka, Amazon Kinesis or Azure Stream), real-time data can be sampled and used to re-train the model making it more robust. Real time data is analyzed by the model at every step and the output is the expected to show the remaining lifespan. Some decision-making criteria can be added on top of the model, like threshold analysis to determine at every step if there are malfunctions in the sensors or unusual values in the readings.

Display Interface

The display interface contains all the relevant information for monitoring. Common visualizations are the equipment wear, anomalies and machine condition.

Conclusion

A considerable amount of data is required to make the best use of predictive maintenance or other advanced analytics. Companies that use data for tracking are the main candidates for predictive maintenance investment. Having the right know-how to define a data model, the best system that would fit a company’s needs and the relevant resources are critical to making significant progress in this process. A gradual implementation starting at simple parallel services, such as predictive maintenance, followed by more complex and sophisticated analytics implementations is key to understanding and realizing the value & benefits that analytics can bring.

Contact our industry experts at Tenthpin if you would like to learn more.

Stay up to date with the latest #Lifeattenthpin #LifeSciences #Pharma #MedDevices #Biotech #Digitalforlife #Thoughtleadership #Medical Technology #AnimalHealth news by following us on Instagram #LifeAtTenthpin Facebook Tenthpin and our Tenthpin LinkedIn corporate page.