(Editorial note: Since publication of this column, SKF has completed the acquisition of Presenso on October 31, 2019. See the official SKF release.)
Whether an unplanned line stoppage at a plant is catastrophic or simply a waste of time, money and resources, everyone agrees that it would be better to know when the drive, motor, gearbox or substation will fail.
Presenso, a two-year-old company headquartered in Haifa, Israel, wants to give food and beverage processors the ability to plan repairs on their schedules. The new company was cofounded by Dr. David Almagor (chairman), Eitan Vesely (CEO) and Deddy Lavid (CTO), all with several decades of combined experience in machine learning; big data architectures; complex R&D; hardware; software; mechanical interfaces; and electrical, mechanical and computer engineering.
Presenso is all about tearing down the rules of expensive rule-based systems, which are reactive in nature, and replacing them with cognitive systems, which provide operational intelligence and deep semantic insights—PdM (sometimes known as predictive asset maintenance) at its best.
Presenso’s system is hardware agnostic, runs on a cloud platform and uses the latest machine learning and big data technologies. Its analytic engine has deep learning capabilities and runs off data accumulated from plant equipment, sensors and controllers. FE recently spoke with CEO Eitan Vesely to learn more about Presenso.
FE: How did Presenso get started?
Eitan Vesely: It starts with a story. I am a mechanical engineer by training and was working as a hardware specialist/support engineer at Applied Materials. Most of the time, I would be sent to production facilities to fix failed machines and bring them back up to production. I would always start by reviewing the historical data and trying to identify the root cause of a machine malfunction. Machines generate terabytes of data, but most of it is not used. Data is simply stored and accessed when something goes wrong.
One day in early 2013, I was sitting at the site of a major manufacturing plant working through the data. The idea came to me that there must be a way to use all this data to predict a machine breakdown before it occurs.
At the same time, my friend Deddy Lavid was working on predictive analytics for the energy sector. We brainstormed with people from the industrial domain who validated the need for a predictive solution for identifying the root cause of failure before machine degradation occurs. That is how our IIoT solution germinated.
Deddy joined me as co-founder of Presenso, along with serial entrepreneur, Dr. David Almagor, and the rest is history.
FE: What does the architecture of a Presenso system look like?
Vesely: Our solution uses automated machine learning (also known as meta-learning). As such, it is agnostic to the sensors and the machines it is monitoring.
Since we offer a SaaS model, the Presenso architecture is mostly cloud based. There is no need for any purchase and installation of new hardware. The Presenso streamer extracts data from SCADA systems. In our analytics platform, data is pre-processed, and we then perform automated machine learning algorithm selection. Production facilities access the Reporting/Business Intelligence Interface to view the health of their machine assets in real time.
FE: How does the system work?
Vesely: We extract operational sensor data from our cloud-based servers. Advanced AI algorithms are used to detect anomalous sensor behavior and also patterns of anomalous behavior. This pattern recognition is important, because it gives insight into the possible root cause of failure.
Automatic alerts are sent to factory technicians for them to remediate the failure threat. At the same time, the dashboard is useful for management, whether at a plant or headquarters. In this way, the overall health of a facility or company’s production assets can be evaluated in real time.
FE: How is Presenso different from other PdM solutions?
Vesely: At a high level, our primary differentiator is that we use existing hardware and do not require the input or training of production facility technicians. Each alternative PdM solution is hardware or resource dependent.
Let’s start with some of the hardware-based solutions, such as vibration or acoustic monitoring. First, they all require the installation of new sensors. In addition, these are point solutions that can identify specific problems, but do not provide a holistic view of a production line.
There are manual statistical modeling packages that require some level of local expertise. The problem with this approach is that statistical modeling is time consuming and does not provide real-time analysis. Given all the advancements in machine learning, a surprisingly high number of facilities are still using Microsoft Excel for their statistical modeling.
Another alternative in the marketplace is the so-called “digital twin.” A virtual clone of the machine is created using the physical blueprint. From my understanding, the digital twin has not had much traction in the food industry, primarily due to the high cost associated with implementation and the significant investment in time by facility technicians to “train” the digital twin on the underlying asset.
FE: How does IIoT fit into Presenso?
Vesely: We are a pure IIoT predictive maintenance solution. We process and analyze the operational data in our cloud, so that there is no need for this activity to occur at a facility level. The production plant gets what it needs the most: alerts to evolving asset degradation and failure.
FE: What additional sensor requirements are there to make Presenso work?
Vesely: We provide a software solution that is based on analysis of operational data generated by sensors. In most cases, there is no need to install additional sensors, as they already exist. We often find that there is a disconnect between centralized information technology (IT) and local operational technology (OT) groups. Where this occurs, there may be challenges accessing data that is already captured.
Let me put this another way: Instead of more sensors, we need a better alignment between the various stakeholders.
FE: Why the food industry? What size businesses could afford Presenso?
Vesely: The driving force here is economics.
The food industry has lagged in the adoption of Industry 4.0 and digitalization. The prevalent approach to asset maintenance is reactive, and run-to-failure is the default. Although it is difficult to get accurate data, the cost of downtime in the food processing industry is estimated at $30,000 per hour. That’s just an average. Depending on the production line, the true downtime cost can exceed $70,000 or even $100,000 an hour, and the annual downtime can exceed 400 hours a year. No matter which estimate you choose to use, annual downtime costs can amount to millions of dollars per factory.
Because we provide a software as a solution service offering, Presenso is affordable to even medium-size production facilities, especially those lacking internal machine learning resources.
For more information, visit www.Presenso.com