IoT Sensor Solutions
We help you choose the right sensor for the use case. We have experience with use of vibration, pressure, thermal IR, and vision sensors.
Best Platform Selection
We have experience selecting and deploying different platforms and can help you choose the best of the best.
Best Use Case Selection
Selection is a tradeoff of failure detectability, cost of failure on downtime and frequency of failure. Assessing detectability before the PoC is the key decision.
The GENESIS Technology Package Dashboard
Solves Problem of High False Positives
Overcomes challenge of limited and/or poor quality data which creates unacceptable false positives and missed true negatives. It does so through a human-in-the-loop approach that incorporates human expertise or inference. It allows existing SMEs to cover more ground and think smarter.
Works for Any Operational Process
Ideally suited for predicting problems that develop over time including degradation in production process, asset health and inventory levels. Diagnostics included.
Gets Enthusiastic User Adoption
Decision makers get free of time-consuming, repetitive tasks and instead are engaged in making more decisions with the data they have been using. A great way to scale your best human talent into more engaged decision making.
Works on Top of Any Cloud Platform
Our alert dashboard and data science currently operates on the AWS platform but can be adapted to operate on any cloud platform. It uses the REACT UX environment, and the data science and dev code is Python.
Our approach to AI workflow was born from the challenging world of the Industrial IoT – enterprise manufacturing in large scale deployments. Here, data is limited in scope and history, and users skeptical and resistant to adoption.
After successful POCs that demonstrated high potential ROI, the deployment got off to a slow start. We found there are many types of failures not related to actual use case that created false positives. Users would not tolerate the workload created by false positives or missing true negatives. Also, we averaged an average true positive prediction time of 30 days! This, too was a problem as users could not see the problem so early. They could only see 3-5 days in advance. So this made them skeptical.
We solved this by developing a failure alert management system where all data relevant to the prediction, even if not used in the ML models, was provided so they could use their experience, their inference to determine an action on the alert. All this served to keep the technician squarely in control by enabling them to approve or disapprove the alert. This approach is so efficient that it enabled one technician to support over a thousand real time vibration sensors across multiple sites. The inference of one technician in the decision making loop made all the difference. And the technicians enjoyed their job by not having the drudgery of making all the calculations, by not having to operate in the fear of missing a failure.
Our approach is commonly known as "Human in the Loop." We use human inference to finalize decisions, teach the models. This overcomes the problem of limited data that does not fully describe the problem. It also prevents alerts that are anomalies not related to the use case, eg transients in operation, startups and shut downs, etc.. More on Human in the Loop...
This GENESIS approach is available on your cloud platform as part of our Technology Package
GENESIS Dashboard Operation
Platform screen shots below from an actual deployment (using anonymized data) show the five steps of operation. It's a strong and common sense workflow integration where users do not need more than 10 minutes training and no data science skills are required to understand best decisions (excepting step #5).
List all alerts and their decision status
Not only does it list all open list, but allows users to place an advance alert prediction into "Tracking" mode so that it can be scheduled for intervention at the best future date.
See underlying data that explains failure causality
AI models predict failures by processing operational data. The operational variables that triggered the alert are shown in this screen. Users obtain a deeper understanding by see the data they have been using for years. This enables them to confirm, monitor or deny the alert.
Classic engineering methods reveal hidden patterns in data
In this case hidden features bearing vibration data provide more precise data for ML models. Applicable to any data including thermal and imaging. Also an excellent source of diagnostics.
Supervisory data is matched to new fault signatures
In this case maintenance work orders that match the fault signature are presented to the user. An excellent form of diagnostics.
Revolutionary Composite Health Index algorithm
Our AI platform is able to composite multiple predictive models to create a more robust predictive alert. This screen is normally hidden from line operators. No data science skills are required to make best decisions.
We are Cloud Platform Neutral
Works with any cloud platform including those below.
Amazon Web Services
The most powerful, scalable on-demand cloud computing platform
The ultimate platform for cloud storage, computing and overall solution building. Offering over 500 libraries for code integration it provides the most power available but also the highest bar on skills required. More from AWS...
Deliver advanced analytics using the latest techniques at scale
A low code data science platform with the power for AI deployment at scale. A centralized, controlled and managed environment offering much of the power of AWS without the code requirements. More from Dataiku...
Depth for data scientists,
simplified for everyone else
A low code, no code data science platform with a remarkable user interface that enables users at all skills levels. It unifies data prep, machine learning, and model operations. More from RapidMiner...
Key Enabling Partners to Stay on Front End of AI Developments