4 Ways Machine Learning Can Save Money for Industry
Machine learning is the boat that’s lifting all tides, providing solutions to problems once thought unsolvable. It’s already being used in financial services, travel, entertainment, and media companies all over the world, generating millions of dollars each year. However, not all machine learning platforms are created equal. In this article, you’ll learn how machine learning can save time and money for your industrial business.
Predictive Maintenance and Mechanical Failure
Standard process control systems often fail to detect minute variations in vibration, sound, and other inputs that can foreshadow mechanical failure. In the past, machine learning for predictive maintenance had been used to augment “root cause analysis” in retrospect — process engineers would share findings and make forward-looking recommendations to engineers and operators.
Today, plants can leverage both real-time and batch analysis, meaning operators will finally be able to benefit from alerts and recommendations with enough lead time to avoid costly trips or unscheduled maintenance. Rotating machinery teams are not only able to create and deploy ML models for failure detection efficiently, they are also able to track these models for drift and retrain them effortlessly, keeping models up-to-date and at maximum efficiency.
Load Distribution
While wind and solar intermittency has only started to stress the physical limits of some electric grids, requirements for future intermittent assets far exceed existing infrastructure capacity. Better predictive balancing, efficient energy storage, and better integration intelligence will be crucial for future grid reliability.
In addition, connecting the consumer to the electric markets will exponentially increase the complexity of system management, commercialization, and optimization. To commercialize the consumer side of the market requires the digitization of a vastly broader set of input/output categories than the wholesale market.
There is no precedent for any human-run balancing at this level of complexity — only real-time machine learning models that can ingest, store and analyze petabytes of multi-domain data will enable grid and utility operators and their AI-based advisory systems to keep up with rapidly changing multi-domain data.
Sensor Failure
Many plant trips are the result of faulty sensors. Some trips can cost millions of dollars a day. Industry-wide, this causes a tremendous amount of productivity loss. These faults often happen because a sensor value is suddenly outside an operating range or not registering. In reality, it’s often the sensor itself that is faulty, and any seasoned operator can distinguish that from the wealth of other data points available. However, operators find it difficult to address these issues in real-time because of the hundreds of other alarms they are simultaneously addressing.
Modern machine learning technology enables operators to run real-time predictions on identified “critical tags” by analyzing up to ten million data-points per second. This allows for a “gut check” of sensor health and allows both operators and maintenance to intervene at optimal times and avoid costly interruptions.
Leak Detection
Leak Detection packages are an integral part of most control systems in the fluid process industries. However, leaks still occur all too often. In addition to the environmental and safety costs leaks impose on society, leaks are often commercially devastating. Early detection in leaks, or better yet, prevention of leaks, is top priority. Basic calculations such as sudden changes in pressure or flow can trigger programmed alarms — but by this point, the damage has already been done.
However, machine learning can identify several digital signatures in various, often disparate data sets that can indicate when a leak may be imminent. This fusion of multi-domain data allows for effective and dynamic ML learning models to be developed, deployed, and improved upon to best predict and avoid leaks.
Solutions
Now that you know what machine learning can do for you, how do you choose between the huge variety of platforms on the market?
Traditional approaches to the application of AI/ML to industrial problems fall into two basic approaches: the Do-it-Yourself assembly of open-source or native cloud software components, and proprietary Black-Box solutions. Both of these approaches have their own benefits and drawbacks.
The DIY assembly of components gives you full control of your platform, but at the cost of engineering complexity and data latency. To provide an efficient end-to-end solution, a large team of distributed engineers must continuously integrate many components. This is expensive, time-consuming, and difficult to execute.
Even if you could find the experts and afford to pay for the DIY approach, the loosely-coupled components need to share data, and data moves between processing engines slowly. In the real-time environment of the industrial world, this slowness could impair a system’s ability to make a prediction of an outage or performance issue with enough time to remediate it.
The black-box proprietary systems have a different drawback: these engineered systems are not open source and require specialized skills to operate. Moreover, they are typically bundled with significant professional services that are notoriously difficult to end. If you sign on, you’re stuck indefinitely. Additionally, some vendors require you to share ownership of your data with them. That asset is yours. Why share it with your competitors?
There is too much data in industrial settings to have a loosely coupled system, and too much at stake to sign your rights away. So choose the best of both worlds: choose Livewire.
The Livewire Operational AI Platform
Livewire is a completely integrated system that is built on open source standards in an open source framework that you can easily try in a cloud-based sandbox. With Livewire, your teams’ resources are spent on building predictive models that perform, instead of patching low-level components. You can rest assured that your staff will not be backed into a corner by opaque, proprietary solutions. Livewire is open source, transparent, easy, and fast.
The Livewire Operational AI platform has three components:
- Connectivity and Ingestion — Tools and APIs to integrate and ingest data from DCS’s, SCADA’s, historians, ERP systems, MES systems and other data sources.
- Prediction Platform — An end-to-end Machine Learning platform for developing, experimenting, and deploying machine learning models.
- Observability Platform — Visualization and alerting tools to surface temporal data, static data, and predictions.
Livewire enables teams of data engineers, operators, and data scientists to work together with unprecedented speed and agility. By using an integrated platform, these teams can deploy machine learning models 100x faster with half the staff.
Don’t believe us? Give it a try! Have your engineers and developers fire up a trial at livewire.ai/trial with their own sandbox immediately.
