Improve Your OEE: Avoid Unplanned Downtimes and Quality Fluctuations
Get Transparency For Your Processes
Why Advanced Process Analytics Matter
Understand to Undertake
Get the Big Picture
Enable personnel to make the right decision at the right time. Most larger issues have complex causes that cannot be eliminated if the interactions are not clear. Empower your staff to see, understand, and act.
Full Coverage for Full Impact
Focus on What Is Relevant
Combining human and machine intelligence allows to make progress in adverse conditions. While AI keeps an eye on everything in the background, you can focus on what matters. With a unique methodology algorithms continuously and instantly tune themselves to your needs.
What is the Challenge?
Getting Things Off the Ground
Getting started is often the hardest part. With over a decade of experience in different industries such as nuclear power or aviation, we offer a modular and proven roadmap to get from planning to production in six weeks.
The focus is on a minimal need for IT resources while obtaining the highest possible security level using capsulation and standardization. Depending on your goals and requirements, we have four different tailored service offerings.
Expert Cloud Service
Caverion Intelligence’s success is based on the combination of process experts from our long-term partner Caverion with our team of data scientists. Fusing deep industry knowledge from a technical and a mathematical side ensures optimal performance around the clock. Besides regular check-ins, you will receive detailed information on how to proceed should there be any louming problems. Anything from what work orders to create, to which additional checks to run, up to how to improve your operations. No need to strain your valuable resources.
Guided Cloud Service
Combine Your Expertise with Machine Learning
Take a deep-dive into specific issues
1. Definition & IT Security Phase
2. Modelling and Setup Phase
3. Calibration & Value Creation Phase
4. Long-term Operation
Standalone UI or deep integration
Either get started with our streamlined UI build on web technologies, integrate UI elements into your tool landscape, or build upon our API.
Success Case Predictive Maintenance Pharmaceuticals
Success Case Predictive Maintenance Bio Fuel
Success Case Predictive Maintenance Gas Turbine
Success Case Predictive Maintenance Coal Power
Success Case Predictive Maintenance Metals
Success Case Predictive Maintenance Nuclear Power
Success Case Predictive Maintenance Aviation
Success Case Predictive Maintenance Buildings
Success Case Predictive Maintenance District Heating
Preventing Disruptions Seamlessly
Operations and Maintenance receive real-time information about issues and take action immediately.
All process areas and all types of potential disruptions stay under control. From failures of individual components using Predictive Maintenance to complex quality fluctuations using advanced AI, Prexello has you covered.
Essentially, the Control Tower team tackles issues before they escalate into problems with two ways of implementation.
Firstly, with our Guided Cloud Service your process expert service have full control and get assistance from our experts when needed.
Secondly, utilizing our Expert Cloud Services as in Caverion Intelligence all worries are taken off your shoulders. You get Results-as-a-Service.
What Customers Say About Caverion Intelligence
From the press release: “The team identified good practices and good performances that will be shared with the nuclear industry globally, including […] a machine learning software to monitor turbine performance”.
“By using the Caverion Intelligence AI solution we have been able to detect and mitigate several deviations long before they escalated into real failures.“
SSAB has a leading position in high-strength steels and related services. SSAB’s Hämeenlinna steelworks produces advanced high-strength galvanised steel for customers in the automotive industry.[…] By optimising the production process, the Hämeenlinna steelworks wanted to minimise the number and duration of disturbances across the galvanising line and to ensure the highest product quality. This was accomplished through the Caverion Intelligence Anomaly Detection service, which uses existing process data in combination with advanced machine learning for early detection of anomalies across the whole process.
“We have integrated machine learning as a part of our reliability engineering process. The results we have achieved are very encouraging and we are expanding the use of this capability in various parts of our processes. Caverion has been an excellent partner for us on this journey. They have a strong, mature and proven service concept and a very competent team. The service is constantly evolving to meet our evolving needs.”
Stockholm’s energy company Stockholm Exergi has been using the Caverion Intelligence Anomaly Detection Solution to monitor its heat and power generation in the greater Stockholm area since 2019. The system supports operators and maintenance personnel and ensures reliable production performance. It also prevents unplanned downtime by detecting anomalies at an early stage. The system does this by continuously analysing thousands of signals and using advanced machine learning algorithms to recognise deviations. By identifying deviations early, errors can be corrected in good time and the risk of downtime in production can be avoided.
A Technical Deep Dive
- Every site, plant and device is unique.
- Physically identical components might come from the same production line in the same batch. After a few months every component has its own history, its own story to tell and the data it is emitting will look different.
- Failures will have different patterns specific to each machine.
- AI must be able to adapt to this and at the same time still be able to transfer knowledge from one component to another.
- Prexello abstracts the behavior creating a robust and invariant description of faults
- Benefitting from AI is a continuous process
- An AI solution should pay for itself within a very short period, weeks or even days, but after that still should continue to deliver more and more value
- A journey that gradually guides you towards a best possible reliability and efficiency
- At the beginning of our AI journey it’s the goal to eliminate big issues such as major unplanned downtimes and then slowly address more and more sensitive and specific issues such as efficiency degradation
- There might be 100,000s of sensors at a site. Often it is not known what these tags mean. Only a small amount of sensor data is being followed up on a regular basis
- When leveraging data-driven AI for OEE improvements this doesn’t need to change
- Focus shouldn’t be on an endless categorization and documentation of tags
- Only if a tag shows that there is a problem you should investigate what that means
- Prexello will assist you with context and guide you to what is important
- –> So, all models will work great from the start
- Most big failures almost never repeat themselves
- For repeating issues, AI rules are created leading to prescriptive detection
- Identifying novel issue before they grow into series problems is hard
- Prexello captures novel issues early on
- Our AI covers all sensors, combining all available information for maximum protection
- –> Focus on things that are causing problems not on things that might cause problems
- Conditions change all the time.
- Hot summers or cold winters, heavy loads or partial loads, freshly maintained or close to the end of its useful lifetime.
- In order for PdM work successfully the AI needs to learn the behavior of all these conditions and must be sensitive to issues in all these situations.
- We have devoted over 10 years of research to solve this problem.
- Lean data connections and data storage solutions can be sufficient
- Target a straight forward collection of data from different data source
- Start lean, learn what is important and can deliver value
- Measurements that were deemed extremely important sometimes contain information that was so trivial it is not needed for predictions
- Take data directly from machines and automation systems
- We focus on a simple, quick approach that lives up to the highest safety standards
- It is clear that machine and process data needs to leverage to improve OEE. Also, this is the only way to implement a state-based maintenance strategy
- On the other side, PdM AI has to integrate seamlessly into existing workflows
- There is already an existing tool landscape and adopting PdM software cannot mandate
- Our PdM solutions has a clear mission that should be performed in the background unless something is wrong
- I.E. it is not the job of a power plant and their personal to run AI, its job is to reliably and efficiently provide electric power
We integrate into maintenance systems and our customizable dashboards can be added to existing tools
- Industrial plants usually provide plenty of data. However, these are often incomplete and not designed for advanced analysis
- Events are not labeled or at least not in quality expected for ML
- Documentation is not focused on data science, e.g. by the minute recording of component replacements or maintenance
- Maintenance logs might contain the days or rough areas in the process where changes happened
- To make standard AI work, data needs to be prepared by a mix of engineers and data science experts
- A practical AI needs to take all this into account and must be robust
- Prexello works with real-life data and quickly learns from your experts
- Launch with imperfect data, get results immediately, quickly reaching highest prediction performance
- Our data scientists monitor and manage everything in the background
- AI project need to ramp up quickly
- Avoid long and unneeded start-up projects
- Start working with the data you have
- Minimize assumptions and modelling for areas that might be completely irrelevant
- Once data is accessible it is the AI’s job to indicate what is relevant and what not
- The roll-out shouldn’t take longer than a few days
- Don’t lose time analyzing processes at your site that run stably and don’t cause any issues
- Our AI will guide you to things that might be relevant and then quickly learn from feedback
What is Prexello GR?
GoldenRun (GR) is relevant in all industries that do processing in batches and that face relevant cost through scraps or rework.
GR focuses on systematically and continuously improving quality & performance factors of OEE.
What Does GR Do?
Mathematical models monitor the ongoing production step by step and alert personal on issues in process or control.
Impact of GR
Increase quality, boost output, and reduce scrap
by continuous learning and improvement through AI
Select a few good batches, if known the best ones.
- Data won’t show the ideal behavior, rather reality creates a distorted image of it –> AI identifies patterns and builds a stochastic model.
- Algorithms extracts an idealized picture of the batch processing, but considers realistic variations.
- Model learns sequence of individual steps and properties of each step.
- Tolerances for each measurement and other properties, such as valid durations, are derived for batches and individual steps in the batch.
Base model is then used for online and offline evaluations. User can give feedback to predictions refining the model on the go.
Models improve over time and learns what is relevant to the user.
Inside the Technology: HiFi Zooming
In which industries does the use of AI bring the greatest success?
In general, AI is suitable for monitoring and improving plants and processes in any industry. Currently, our solutions are used in power generation (cogeneration, nuclear, waste, hydro), pharmaceutical, metal, chemical, aviation, pulp and paper. The crucial point is the availability of continuous process data on which the AI is built.
Do I need to purchase new sensors or other hardware?
Our solutions uses existing data recordings or sensor values. These form the basis for the implementation. Data can be obtained and merged from all relevant sources. In the course of projects it may turn out that additional measuring points are advantageous, but this is only a second and optional step.
How is data security ensured?
All data is processed either in our own data center or in a private cloud with ISO 27001 certification. Either in Germany or France.
All processes are specially designed for industrial applications to satisfy the highest safety requirements. In cooperation with customers, all safety measures are regularly reviewed.
How long does the implementation take?
Usually, there are only a few weeks between the start of the project and the full availability of the solution. As soon as the data exchange has started, the first results are available within a few days.
The roll-out process follows an established procedure. First, the data is structured together with the customer and a data interface is created. Then we start with the mathematical modeling and user trainings begin in parallel. During the first weeks, our experts provide close support.
How is the data transferred?
The actual solution is highly dependent on the industry, but there is a proven, robust solution for every situation. Data can be transferred as a comma separated file (CSV) or via a protocol or procedure such as MQTT, OPC-UA, Rest API etc. A connection to a cloud storage is also possible.
Additionally, we have a network of cooperation partners who provide solutions for special cases and can also establish a secure and stable data transfer in complicated cases.
What resources and skills do we need for a successful deployment?
We handle the setup of the AI based on your data. For this purpose we conduct a workshop to clarify the crucial points. We then support you in rolling out the new capabilities in your company through training and regular online meetings. After a short learning phase, the time required to operate the web interface takes about 15 to 30 minutes daily. During this time, warnings and hints of the AI are reviewed and evaluated.
No special mathematical knowledge or experience with AI is required.
How can we best approach AI and predictive maintenance as topic in our company?
Based on existing data, important insights can often be obtained without creating an interface for online data. Problems from the past can be analyzed and the first improvements can be implemented in processes.
All in all, our AI is structured in such a way that it delivers good results with only a little data. This also makes it easy to achieve continuous data analysis in a timely manner without the need for real-time data connection.
Louis von Beaulieu
Would you like to learn more on how your company can benefit from the application of mathematical technologies and what the benefits you can expect in your specific case?
Telefon: +49 160 90 83 97 56