It’s a timeless manufacturing goal: to produce high-quality products at a minimum cost. Factory 4.0 is already demonstrating its value by enabling manufacturers to reach this goal more successfully than ever, and one of the core technologies driving this new wave of ultra-automation is Industrial AI and Machine Learning.

Data has become a valuable resource, and it’s cheaper than ever to capture and store. Through the use of artificial intelligence, specifically Machine Learning, manufacturers can use data to significantly impact their bottom line by greatly improving efficiency, employee safety, and product quality.

 

Powering Predictive Maintenance with Machine Learning

Maintenance represents a significant part of any manufacturing operation’s expenses. For this reason, Predictive Maintenance has become a common goal amongst manufacturers, drawn by its many benefits, with significant cuts in maintenance costs being one of the most compelling.

While certain manufacturers do perform Predictive Maintenance, this has traditionally been done using SCADA systems set up with human-coded thresholds, alert rules, and configurations.

This semi-manual approach doesn’t take into account the more complex dynamic behavioral patterns of the machinery, or the contextual data relating to the manufacturing process at large. For example, a sensor on a production machine may pick up a sudden rise in temperature. A static rule-based system would not take into account the fact that the machine is undergoing sterilization, and would proceed to trigger a false-positive alert.

In contrast, Machine Learning algorithms are fed OT data (from the production floor: sensors, PLCs, historians, SCADA), IT data (contextual data: ERP, quality, MES, etc.), and manufacturing process information describing the synchronicity between the machines and the rate of production flow.

In AI, the process is known as “training”, enables the ML algorithms to detect anomalies and test correlations while searching for patterns across the various data feeds.

The power of Machine Learning lies in its capacity to analyze very large amounts of data in real-time, and propose actionable responses to issues that may arise. The health and behavior of every asset and system are constantly evaluated and component deterioration is identified prior to malfunction.

 

Enabling Predictive Quality Analytics with Machine Learning

Preventing downtime is not the only goal that industrial AI can assist us with. The quality of output is crucial and product quality deterioration can also be predicted using Machine Learning. Knowing beforehand that the quality of products being manufactured is destined to drop prevents the wastage of raw materials and valuable production time.

 

Supervised & Unsupervised Machine Learning

Machine Learning can be split into two main techniques – Supervised and Unsupervised machine learning.

Supervised Machine Learning

In manufacturing use cases, supervised machine learning is the most commonly used technique since it leads to a predefined target: we have the input data; we have the output data; and we’re looking to map the function that connects the two variables.

Supervised machine learning demands a high level of involvement – data input, data training, defining and choosing algorithms, data visualizations, and so on. The goal is to construct a mapping function with a level of accuracy that allows us to predict outputs when new input data is entered into the system.

Initially, the algorithm is fed from a training dataset, and by working through iterations, continues to improve its performance as it aims to reach the defined output. The learning process is completed when the algorithm reaches an acceptable level of accuracy.

In manufacturing, one of the most powerful use cases for Machine Learning is Predictive Maintenance, which can be performed using two Supervised Learning approaches: Classification and Regression.

These 2 approaches share the same goal: to map a relationship between the input data (from the manufacturing process) and the output data (known possible results such as part failure, overheating etc.)

  • Regression

Regression is used when data exists within a range (eg. temperature, weight), which is often the case when dealing with data collected from sensors.

In manufacturing, regression can be used to calculate an estimate for the Remaining Useful Life (RUL) of an asset. This is a prediction of how many days or cycles we have before the next component/machine/system failure.

For regression, the most commonly used machine learning algorithm is Linear Regression, being fairly quick and simple to implement, with output that is easy to interpret. An example of linear regression would be a system that predicts temperature, since temperature is a continuous value with an estimate that would be simple to train.

Regression labeling for Predictive Maintenance.
Regression labeling for Predictive Maintenance. Each recorded instant (5Y, 4Y, 3Y etc.) describes the Remaining Useful Life of an asset before it is predicted to fail.
  • Classification

When data exists in well-defined categories, Classification can be used. An example of Classification that we’re all familiar with is the email filter algorithm that decides whether an email should be sent to our spam folder, or not. Classification is limited to a boolean value response, but can be very useful since only a small amount of data is needed to achieve a high level of accuracy.

In machine learning, common Classification algorithms include naive Bayes, logistic regression, support vector machines and Artificial Neural Networks.

Predictive Maintenance makes use of multi-class classification since there are multiple possible causes for the failure of a machine or component. These are possible outcomes that are classified as potential equipment issues, calculated using a number of variables including machine health, risk levels and possible reasons for malfunction.

 

Unsupervised Machine Learning

With Supervised machine learning we start off by working from an expected outcome and train the algorithm accordingly. Unsupervised learning is suitable for cases where the outcome is not yet known.

  • Clustering

In some cases, not only will the outcome be unknown to us, but information describing the data will also be lacking (data labels). By creating clusters of input data points that share certain attributes, a Machine Learning algorithm can discover underlying patterns.

Clustering can also be used to reduce noise (irrelevant parameters within the data) when dealing with extremely large numbers of variables.

  • Artificial Neural Networks

In the manufacturing sector, Artificial Neural Networks are proving to be an extremely effective Unsupervised learning tool for a variety of applications including production process simulation and Predictive Quality Analytics.

The basic structure of the Artificial Neural Network is loosely based upon how the human brain processes information using its network of around 100 billion neurons, allowing for extremely complex and versatile problem-solving.

A basic schematic of a feed-forward Artificial Neural Network.
A basic schematic of a feed-forward Artificial Neural Network. Every node in one layer is connected to every node in the next. Hidden layers can be added as required, depending on the complexity of the problem.

This ability to process a large number of parameters through multiple layers makes Artificial Neural Networks very suitable for the variable-rich and constantly changing processes common to manufacturing. Moreover, once properly trained, an Artificial Neural Network can demonstrate a high level of accuracy when creating predictions regarding the mechanical properties of processed products, enabling cuts in the cost of raw materials.

 

Data Preparation

Machine learning is all about data, so understanding some key elements about the quality and type of data needed is extremely important in ensuring accurate results.

With Predictive Maintenance, for example, we’re focused on failure events. Therefore, it makes sense to start by collecting historical data about the machines’ performance and maintenance records in order to form predictions about future failures.

Since the operational lifespan of production machines is usually a number of years, historical data should reach back far enough to properly reflect the machines’ deterioration processes.

Additionally, other static information about the machine/system is also useful such as data about a machine’s features, its mechanical properties, typical usage behavior and environmental operating conditions.

Next, certain questions should be answered to help focus on the data that is most crucial to our needs:

  • What are the various types of failure that can occur with this component/machine/system?
  • Which failure events are we interested in trying to predict?
  • Is the failure a sudden, focused event, or is there a slow decline before complete malfunction?
  • Which components are typically associated with this type of failure?
  • Which parameters should be measured that most signify the state of component/machine health?
  • What is the required accuracy and frequency of the measurements needed?

The questions above should be answered by both domain specialists and data scientists, resulting in the final and most important two questions:

What question do we want the Machine Learning model to answer? And, is it possible to answer this question using the data that’s available?

 

The Groundbreaking Benefits of Machine Learning and AI for Manufacturing

The introduction of AI and Machine Learning to industry represents a sea change with many benefits that can result in advantages well beyond efficiency improvements, opening doors to new business opportunities.

Some of the direct benefits of Machine Learning in manufacturing include…

  • Cost reduction through Predictive Maintenance. PdM leads to less maintenance activity, which means lower labor costs and reduced inventory and materials wastage.
  • Predicting Remaining Useful Life (RUL). Knowing more about the behavior of machines and equipment leads to creating conditions that improve performance while maintaining machine health. Predicting RUL does away with “unpleasant surprises” that cause unplanned downtime.
  • Improved supply chain management through efficient inventory management and a well monitored and synchronized production flow.
  • Improved Quality Control with actionable insights to constantly raise product quality.
  • Improved Human-Robot collaboration improving employee safety conditions and boosting overall efficiency.
  • Consumer-focused manufacturing – being able to respond quickly to changes in the market demand.

Lior is a digital marketing manager with a passion for everything IoT. When he’s not busy writing about industry 4.0 and its benefits for manufacturers, he’s usually sleeping.