Seite wählen

Leider ist der Eintrag nur auf Amerikanisches Englisch verfügbar. Der Inhalt wird unten in einer verfügbaren Sprache angezeigt. Klicken Sie auf den Link, um die aktuelle Sprache zu ändern.

Leider ist der Eintrag nur auf Amerikanisches Englisch verfügbar. Der Inhalt wird unten in einer verfügbaren Sprache angezeigt. Klicken Sie auf den Link, um die aktuelle Sprache zu ändern.

Gartner has indicated that AI and Advanced Machine Learning is the top technology trend for 2017. As the field of Industrial Analytics becomes more prevalent, factory owners are beginning to evaluate how Automated Machine Learning fits into their Smart Factory Roadmap. This blog article explains three layers of Automated (Unsupervised) Machine Learning and how they apply
to machine asset maintenance.

Three Layers of Automated Machine Learning

The basic layer of Automated Machine Learning is Unsupervised Anomaly Detection. An Advanced AI algorithm accesses a factory’s Historian database and builds a computer model for the estimated machine behavior based on various statistical and computational methods. It is called Unsupervised because there is no need for data labels to teach the algorithm expected behavior (supervisory signal). Instead, the algorithm learns the normal behavior of the sensor. Based on this understanding the model can detect abnormal behaviors. The algorithm is also able to approximate a baseline for the number of expected abnormalities for a given period. In this way, false positive alerts are minimized. When deviations from the expected behavior are detected, alerts are triggered as the potential the early-warning trigger of a machine breakdown.

One of the major advantages of Automated Machine Learning is that it does not require human intervention. The algorithm accesses and analyzes vast amounts of data that no human could process. With cloud computing technologies, Machine Learning can scale. For instance, in the case of Presenso, all the sensor assets of a factory can be analyzed concurrently in real-time. Deep learning solutions that require (almost) no human input can also reduce maintenance staff overhead and shift facility resources to other critical tasks.

Not all anomalous sensor data is necessarily indicative of degradation or upcoming failure. For instance, it is possible that instead of a sensor data indicating machine downtime, the sensor itself is sending out faulty data. For that reason, Advanced AI algorithms which are based on highly complex data models is required.

The next level of Autonomous Machine Learning is Correlation Pattern Recognition. The AI algorithm is not only looking for abnormal data patterns but is looking for correlations between abnormal patterns. With Machine Learning, the algorithm takes a broad view by analyzing all the anomalous data and then finding hidden patterns between the different patterns. This way, the algorithm makes a deeper level of analysis, aiming in creating more generic, robust and reliable models.

The third aspect of Automated Machine Learning is Root Cause Analysis whereby the Smart Factory gets a holistic behavioral diagnosis. There are many hidden patterns that only a Machine Learning has the computational power to identify. In a typical factory environment, the cause /effect pairing is typically assigned to intuitive correlations based on human logical. In the Autonomous Machine Learning scenario, the advanced algorithms detect counter-intuitive root cause analysis.

By analyzing anomalous data from all the sensors in a production facility, the algorithm can identify the sequence of abnormal events and trace the root cause of machine failure. Most importantly, by isolating the root cause, the factory can also prioritize troubleshooting.

The Role of Automated Machine Learning in the Smart Factory

Automated Unsupervised Machine Learning can help factory owners accelerate their move to the Smart Manufacturing. With Supervised Machine Learning, the algorithm needs to “learn” the asset. For some solutions, such as Siemens MindSphere, an exact virtual replica is required so that the algorithm can learn the exact specifications of the machine. The only way to determine whether a spike in pressure or velocity is meaningful is for the algorithm to understand the machine intimately.

Unfortunately, this knowledge comes with a steep price both in terms of technology and resources.

With Automated Unsupervised Machine Learning, the algorithm does not need to “learn” the intricacies of the machine because it is trained to detect real time deviations from normal data behavior. It looks for patterns that would ordinarily be overlooked in the traditional Supervised model. Using the example of a medical diagnosis, the algorithm would be able to perform a holistic evaluation of the patient’s physiology and combine both traditional and alternative medical tests.

Let’s return to the topic of Industrial Analytics and the Smart Factory. In the case of Unsupervised Machine Learning, the algorithm is agnostic to sensor and asset class or age. The algorithm is simply detecting abnormalities in the sensor data behavior to provide advanced notice of impending degradation or machine failure.

If this sounds too good to be true, that’s because until recently data scientists such as myself were not focused on applying the latest advances in Machine Learning to the industrial domain. Because Presenso’s solution is cloud-based and does not require operational support at the facility level, an industrial plant can get a real-time view of the performance of all its factory assets.