loading

Use Cases

Virtualitics Immersive Platform (VIP) and Virtualitics Predict use cases

team wokring

Use cases by:

Use cases by:

Summary

Virtualitics Immersive Platform (VIP) unlocks AI for Industry 4.0 Asset Productivity
Overview

The immense amount of data available today creates both opportunities and challenges. Whether it’s data acquisition, preprocessing, analytics, or general interpretation, organizations can see the transformation but do not feel entirely prepared to handle this on their own. Incorporating AI to make predictions requires deep experience in the end-to-end pipeline of developing and managing predictive models. Beyond this, a black-box solution does not cut it anymore. Organizations are looking for solutions that are highly accessible and can be easily understood. In order to accomplish this, organizations need different forms of communication in order to drive the information gaps they experience today that impede their most mission critical business decisions.

Virtualitics helps organizations in manufacturing, oil & gas, energy/utilities, natural resource, aerospace, telecommunications, and transportation/logistics to quickly analyze sensor data in order to determine the condition and performance of their mission critical equipment.

Applications
Virtualitics Immersive Platform (VIP) for Industry 4.0

Most solutions require heavy hands-on custom work that takes months to implement. This limits teams from understanding why mission critical equipment might be performing or behaving in a certain way today. VIP enables anyone to instantly jump into data from sensors, SCADA and other asset management systems and quickly characterize why equipment might be behaving a certain way using our no-code embedded AI routines, multi-dimensional visualizations, and 3D digital objects. Connect both your entire ecosystem of data and personnel to improve understanding of data, alerts, and predictive models.

Virtualitics pyVIP for Industry 4.0

Virtualitics pyVIP is our powerful python API that enables flexible integration with tools and systems in your ecosystem. The API enables a Data Scientist to visualize any of their predictive models to enhance model explainability, performance, and overall communication to stakeholders. The API creates the common language bridge and dramatically improves any organizations communication and integration of predictive insights.

Use Cases

Missing the ‘Why?’ in Industry 4.0

Some of the challenges in Asset Productivity and how Virtualitics helps:

Challenges
  • Sensor data informs and alerts but does not characterize why an event might be happening.
  • False positives and lack of understanding lead to muting of alerts, which eliminates the impact of prescriptive maintenance.
  • Professionals using IoT data are not all built the same and have siloed operational knowledge.
  • Interact, explore and ask questions of your data in a natural, effortless way.
  • Maximize performance by visualizing interconnected relationships within the data.
  • Collaborate with team members throughout the organization, external partners, and field services personnel to discover and deploy improvements.
  • Reduce costs by identifying unknown bottlenecks, inadvertent errors and constraints.
  • Drive growth by unlocking new resources, layers, and patterns in your network of services, distribution, logistics, supply chain, and channels.
  • Monitor and control daily operations and logistics across multiple dimensions.

Solution

VIP’s no-code embedded AI and multi-dimensional visualizations enables analysts to quickly identify “why” something is happening.

Case Study

Virtualitics improves energy production of wind turbines
  • Anomaly-detection algorithms revealed the root cause of high temperature operational malfunctions which required excessive servicing, costing more than $500,000 per year in downtime and energy losses.
  • Interactive data exploration in 3D empowered wind farm engineers, analysts and operators to visualize and identify the key driver for underperformance: the OEM designed cooling system as the culprit, which was underpowered relative to the wind turbine designer’s recommendations.
  • AI routines highlighted drivers of energy production and identified ideal operational conditions with respect to auxiliary power.
  • The wind farm’s owner operator commented: “The visual insights derived from VIP would have enabled our team to work with the manufacturer to upgrade the cooling system sooner and at relatively low cost, saving millions of dollars in lost production, warranty costs, insurance, spare parts and equipment expenses.”

A leading multinational company providing energy and automation solutions for efficiency and sustainability looked for improvements in both identifying and understanding equipment failures. The company’s businesses are composed of multiple physical and digital assets that include wind and solar power, energy storage, backup critical power, cooling, and industrial automation controls. These assets support some of the world’s most mission critical facilities. The large number of components in complex industrial systems can fail at any time, due to a variety of causes ranging from grid disruptions to mechanical and electrical faults, and random weather events. As shown in this whitepaper, monitoring, detecting and predicting the root causes of asset failure and underperformance is essential for Industry 4.0 operations. The customer applied Virtualitics Immersive Platform to easily map more than 150 IoT data parameters and more than 50 million data points across two sites to diagnose, identify and visualize the key drivers of equipment failure and energy output.

Using Virtualitics Immersive Platform (VIP) to characterize equipment failure

The company wanted to analyze two industrial wind turbine sites located in different facilities more than 60 miles (100 Kilometers) apart. The turbine’s performance was captured through 150 individual IoT sensors controlled by a PLC system. Operators and OEM Engineers monitored these facilities through a SCADA system with built in time series historian views and automated alerts and alarms. These alerts notified maintenance personnel of events such as operation errors, software malfunctions, grid outages, equipment breakdowns, or abnormal temperatures. The company was aware of some of the big challenges they faced with their existing analytical software tools:

  • Limited capability and time to explore data. Operators and analysts often needed to pull in data science and IT resources in order to perform impactful analysis.
  • These time consuming analyses required the use of expensive external consultants and software coding experts who performed the analysis on difficult to comprehend tools and blackbox models.
  • Solid knowledge and understanding of the systems often led to fragmented decision making whose impact varied in how it was tracked or even measured. Operators and engineers could not visualize the complete set of interactions that most affected the key performance metrics.
  • False alarms created a narrative where you couldn’t always trust the time-based alerts or alarms driven by standard predictive models.
  • The company wasted significant resources not being able to properly assess and proactively react to unexpected deviations in performance of their mission critical assets. Unscheduled maintenance and costly downtime events resulted in more than $5 million of unplanned expenses over 10 years due to a lack of actionable insights connecting human intelligence and vast amounts of IoT data.
  • The company’s multinational team and the OEM’s factory engineers based in Europe and Asia could not easily communicate or share insights based on a unified view of asset-specific advanced analytics models.
VIP’s powerful technology meets you where you operate best

VIP’s powerful visualizations enabled the company to look at data in more ways than one. Whereas traditional SCADA, Process Data Historians, and Asset Performance Management tools focus on informing with 2D line charts, data tables, and pre-set alarms and alerts, VIP delivers insights in a visual, collaborative and intuitive manner. VIP understands that the story is often not as accessible or easy to communicate to your entire audience or extended set of collaborators.

Equipment operators at the organization have an intimate knowledge of their systems and equipment, and must respond rapidly to demanding and unexpected changes in the process and physical environment. Manufacturers have subject matter expertise in specific subdomains such as the electrical, mechanical, and thermal subcomponents of the assets but lack a transversal understanding across geographical sites and a limited view of hybrid facilities where there may be equipment from many different manufacturers and vendors in operation. Service technicians and field workers understand how to perform efficient and targeted repairs if the defective conditions are identified in a specific and clear way by the manufacturer or engineering team. Achieving this type of effective team coordination involves analytics beyond analyzing trends.

To solve this challenge, the first step was to render the real-world environment of the wind turbines as a digital twin representation. This approach enabled everyone in the team to link the data exploration with data and enable all disciplines to instantly begin to understand their data.

 

VIP’s Smart Mapping instantly helps you understand why equipment is performing in a certain way

Using one of VIP’s patented AI routines, Smart Mapping, the company was able to immediately characterize what was driving energy production in these two windmills. Smart Mapping is able to rank which features in the dataset were most influential in driving performance of our target variable, Energy Production, and almost instantly suggest the most optimal visualization in order to interpret and explain the insights. With a few button clicks, and in a no-code environment, multiple stakeholders were able to quickly validate or learn something new from their data.

In the image above, the suggested plot uses 5 dimensions to communicate the insights. Our target variable, Energy Production, is on color. We can quickly assess states of high energy production and lower production. Unlike 2D visualizations, we are immediately able to identify areas for exploration (indicated by the annotations in the image). 

 

From the visualization we are able to see and fully experience that the biggest contributors to performance are wind speed and auxiliary power. We also notice that there are two different performance ranges within auxiliary power, conditions in which the wind turbine operates with low auxiliary power draw (regardless of wind speed) and conditions in which the turbine operates with high auxiliary power draw, and simultaneously high bearing temperatures within the nacelle’s gearbox. With regards to bearing temperatures, we learn that it can be a leading indicator of reduced performance when the temperature gets too high and the overall system begins to tap out the maximum auxiliary power range as the cooling system ramps up. While veteran operators at the company understood that bearing temperature was an important metric to monitor, they did not understand that there was a subtle relationship between bearing temperature and auxiliary power that was deterministic of performance. This level of characterization became a necessary tool for the company to create a common language across the mechanical engineering, maintenance, and electrical power teams.

VIP’s Anomaly Detection quickly finds the problem

Using another no-code embedded AI-routine, the company was able to identify two large anomalies; potential malfunctions due to high bearing temperatures and high Line Side Converter (LSC) temperatures. These two subcomponents were initially analyzed in a siloed fashion by the operator, with the problem of bearing temperatures tackled by the maintenance department through more frequent oil replacements, and the problem of LSC temperatures heavily analyzed by the power engineering team to identify disturbances in the grid

In this image, the haloed points represent points that were detected as anomalies. As wind speed increases, energy production remained strangely stagnant for these points. When this happens, it follows that the turbine malfunctioned and required service just a few days later (indicated by the PLC state in the color Red).

By applying the temperature variable on color, we are able to identify a likely culprit of the error state in analyzing the high bearing temperatures (shown in color Red).

Using a similar process to the bearing temperature anomaly, we were able to explore the high Line Side Converter (LSC) anomaly, represented by the large Blue point in the first visualization. In the image below, we’re able to see a point (haloed and color Purple) that indicates high converter temperatures that are causing a PLC error.

Using color to differentiate the range in LSC temperature, we were able to see a sudden spike in converter temperature (indicated by color Red) and a likely reason to cause the error, an underpowered cooling system as designed by the manufacturer which resulted in high bearing temperatures and high Line Side Converter temperatures at critical points in the power curve. Turbine failures occurring at these critical points, in high wind speeds, resulted in severe mechanical and electrical damage to the wind turbines due to sudden hard stops.

Summary
Virtualitics Predict for Airplane Predictive Maintenance
  • Customized solution yields $6-15m annual savings in estimated maintenance costs
  • Integration into existing workflow helps drive action early and leads to 2-3% decrease in overall machinery downtime
  • Ground team experiences 50% reduction in unscheduled maintenance on monitored systems within first three months

A large organization managing a global fleet of aircraft deployed Virtualitics Predict as a Predictive Maintenance solution to improve the uptime rate of their assets. Artificial intelligence-driven predictive models identify parts that are likely to fail and alert the maintenance team with enough lead time to take proactive action.

Virtualitics Predict: How We Engage

Our team worked shoulder to shoulder with the client to determine all the necessary data sources, predictive needs, and reporting requirements. By working so closely with the end users, we were able to deliver a product that meets their exact needs while saving time thanks to our internal development platform for building and deploying our predictive models. 

AI explainability is key to creating buy-in and adoption across an organization, so we ensure that we are providing clear and informative details about how our models are performing as the user is running the tool, not just when we first deploy Predict.

The Virtualitics Predict platform was designed to integrate into existing workflows and alerting systems. In this example, maintenance alerts are automatically passed on to relevant technical teams and sourcing departments using existing workflows. This sped up deployment, removed adoption headwinds and increased how quickly the organization began to see results.

 

Case Study
Virtualitics Predict: The Analysis

There were two main objectives: first, predicting which aircraft were likely to experience downtime in the next month; and second, predicting which parts were most likely to cause each aircraft’s downtime.

In order to perform this analysis, we had several different data sources to leverage. The most robust data source was a database of maintenance logs, which contained information about which parts were repaired on the aircraft, as well as free-text notes from the aircraft maintainers. There were also databases with flight logs and sensor data from the aircraft, which were included to provide more context. Our team decided to also pull in weather data from a historical online data source in order to supplement the client’s own data sources.

The first step in the analysis was to perform some feature engineering to identify the most predictive factors leading to aircraft downtime. We recognized that there was a wealth of information locked up in the free-text maintenance notes, so we applied some natural language processing (NLP) techniques to extract key topics from each set of notes. These advanced analytics capabilities allowed us to achieve a much higher level of accuracy in predicting the aircraft downtime since these extracted features had very high predictive power.

Note: data for illustrative purposes only

We also aggregated and merged the other three data sources to feed them all into our predictive model. The flight logs, weather data, and sensor data were all recorded at different frequencies, so we used different aggregation methods for each data source, utilizing the most relevant information from each. By working with the client, we identified thresholds for critical sensors that would indicate an issue with the aircraft, and we passed this important information to the predictive model.

After merging all of these disparate data sources, the next step was to present several charts and tables to report on the historical situation. This allowed the organization to investigate if there were any general trends in the historical data before examining the predictions, again providing useful context.

The very last intermediate step before presenting the predictions was to provide an entire section dedicated to AI explainability. This was critical for adoption of the machine learning model by ensuring that the organization could consistently examine the performance of the model and understand what it was actually doing. If machine learning models are implemented as a black-box solution, it is difficult to understand what they are doing and trust the results. The added transparency from this section increases confidence in our solution and demonstrates the value added from Predict.

Note: data for illustrative purposes only

Finally, we presented the predictions themselves in a few different formats that were used by management to plan for the next month’s maintenance and by the maintainers to resolve any predicted critical issues right away. For the maintenance analyst, we also integrated Predict with an existing part inventory system so that they could ensure that the necessary replacement parts were in stock, avoiding any additional downtime due to waiting for those parts.

Note: data for illustrative purposes only