Democratising Machine Learning for Operations
Introducing TrendMiner’s Machine Learning Hub
Machine learning projects have long been the realm of central data scientist groups. While use cases for machine learning are plentiful, central data science groups are overloaded with projects. Production sites see improvement possibilities left on the table. What if companies could put the power of machine learning in the hands of all operational experts? And what if they could improve collaboration between the central group and local sites to solve more complex use cases to exceed business goals?
Common Data Science Challenges
Due to their successes, central analytics groups lack the capacity to handle all the data science projects throughout an organisation. Because data scientists are in short supply, they often must prioritise improvements. This can lead to long delays in cycle times of projects for the local production facilities. Smaller use cases fall to lower priorities in a central group. Furthermore, data scientists in a central group speak the language of math and statistics and often do not understand production process behaviour nor contextual data.
Meanwhile, plant operators have limited time for complex use cases. They also lack the knowledge and experience necessary to create and deploy machine learning models. The result is missed improvement opportunities and the underutilisation of operational experts. Frequently, there is no smooth symbiosis between plant-level and central-level stakeholders.
Completely centralised data science groups are responsible for operational improvements. They often have the experience and advanced tooling necessary for more complex machine learning projects. On the other hand, completely decentralised groups place the duty of improving operations in the individual business units. When data management is completely decentralised, the groups have no overall strategy and frequently lack the analytics skills necessary to evaluate operational data.
Often, the best solution is a blended model that uses a federated central group as a facilitator with individual business units responsible for continuous improvement. A state of the art industrial analytics solution reduces the demand on central analytics teams, accelerating the adoption rate of data science projects, and fosters efficient collaboration between operations and the central team. With machine learning democratised to anyone in the organisation, machine learning models can be easily shared, deployed and reused across the organisation to improve operational performance for a wide variety of use cases.
Introducing Machine Learning into Operations
Instead of juggling data in spreadsheets and using the limited analytics capabilities of trend clients, engineers today can work with an advanced analytics solution to analyse, monitor and predict production and asset performance. All sensor generated time-series data can be enriched with contextual data to have a much clearer view on operational performance locked in data residing in other business applications such as maintenance management systems, laboratory information management systems, etc. All the analytics based results to make data-drive decisions can be shown in production cockpits for daily dashboarding.
Some use cases such as anomaly detection models, complex soft sensoring, predictive maintenance, predictive quality, complex sustainability reporting cases and visualisations may require machine learning models to be deployed for operations. In that case the operational experts such as engineers and controllers can prepare analytics based views on production and context data as data frames for the machine learning models.
The data scientist can then use the input for creating and training their machine learning models with use of open source Python libraries. Their results should be easily made available to all production sites and deployed within the industrial analytics solution the engineers are using. The engineers can then use machine learning model tags as if they were normal sensor generated data, for visualisation, analysis and monitoring. Production cockpit dashboards can be extended with notebook cell outputs.
Data scientists and operational experts can visualise notebook cell outputs as tiles in dashboards. They also can generate interactive graphs that can be shown directly on dashboards and show forecasted model outputs on value tiles (such as for predicted maintenance).
Improve Operational Performance with TrendMiner
With TrendMiner’s Machine Learning module, organisations can solve more use cases and prevent performance improvements projects from being left on the table. The solution places machine learning in the hands of every process engineer. The result is cross-functional and global collaboration among operational experts and data scientists; improved employee efficiency for all stakeholders; a higher level of operational excellence, and an increasing competitive edge.
+32 11 263830
info@trendminer.com
TrendMiner
- 003211263830
- http://www.trendminer.com
- Kempische Steenweg 309/5 Corda Building 2 Hasselt B-3500 BE
