Qmerit AI automates home electrification - re-thinking a major infrastructure challenge

New electrical grid approaches are required as the world embraces Net Zero goals. Ultimately, this will be a big, multi-trillion-dollar project requiring thinking about how different kinds of electrical grid infrastructure are designed and built. Now, Qmerit, one of the largest electric equipment suppliers in the US, is hoping to use AI to accelerate some of the challenging aspects of this process within homes and businesses. The new tool uses computer vision and AI to help assess current electrical panels to automate installation plans and file reports. 

Before diving deeper, it’s worth observing that achieving Net Zero goals depends on balancing renewable energy resources like solar and wind with the physical infrastructure or wires for reliably and safely delivering this to homes and businesses. The big deal is that we need to plan how traditionally liquid or gaseous forms of energy can be carried and managed across electrical wires rather than pipes and trucks. The net result is that a lot of new wires and digitalization will have to be installed to account for this transition.

Broadly speaking, there are three types of electrical infrastructure involved. Transmission grids are responsible for getting energy from power plants where energy is cheap to regions with the highest demand across countries and larger areas. Distribution grids relay the energy from transmission grids to local users. Then microgrids and nanogrids manage the last 50-1000 feet to cars, heat pumps, and batteries in offices, factories, apartments, and homes. This is where the new Qmerit technology comes in. 

Most homes and offices were built long before engineers considered the energy demands of electric cars, heat pumps, battery storage and renewable energy. Figuring out how to connect these increasing and bi-directional loads has traditionally been very complex, requiring over 8,000 hours of book and on-the-job training. The new technology promises to bring the same kinds of automation and guidance that GPS and Uber brought to the Taxi industry to microgrid and nanogrid installations.  

Domain-specific AI is key

The new AI was built by analyzing over 269,000 residential and commercial installations in North America. It was trained on thousands of data points about electrical panel pictures, technical characteristics of the installation, and customer outcomes. 

Qmerit worked with Schneider Electric, which has developed an AI hub built by over four hundred data scientists, solving various challenges for the company and partners. It also worked with Microsoft to develop and improve computer vision models for recognizing details of an electrical panel from pictures taken with a smartphone. In addition, Qmerit built a new tool called a Load Capacity Recommendation Engine to interpret the panel's suitability for EV chargers, battery storage, and heat pumps. 

Qmerit CTO Manoj Puthenveetil says the whole project took about ten months, most of which focused on organizing its existing data:

In this scenario, we needed to rely heavily on our domain knowledge to ensure that we labeled the data correctly and trained the model to recognize the most important technical characteristics of the electrical distribution panel that are relevant for a given install, such as the type of breakers, the brand of the panel, connected loads, etc. We designed the whole system to continually improve as we learn more real-world insights in the future. Any changes or adjustments the electrician makes are automatically fed back to the model to create a self-learning loop.

Qmerit also engaged electricians to work with its AI data scientists. Puthenveetil explains: 

This collaboration was critical to ensure we correctly labeled the data to extract features necessary for the model and test out the accuracy of the training. This was a challenging process as we had to learn what to focus on based on an evolving understanding of AI model performance and its alignment with the overall project objective. We took an iterative approach to ensure that we re-calibrated the results with the field engineers to ensure that the results were valuable to the overall customer experience.

The new tools allow skilled electricians to change various inputs detected by the AI tool. These changes are then used for the calculations and subsequent recommendations. Any updates or changes made are logged by the system and automatically considered for future training of the model. This continuous loop ensures the model improves as it interacts with real-world electrifications.

The electrification experience

The result is that consumers and businesses can kick start their electrification journey by filling out a form and taking a picture of their existing power panel. The AI helps reduce the number of questions since it can infer more insights from a picture of the existing power panel. 

Once the request is submitted, the Qmerit system provides an automated estimate for the installation job. The assigned electrician reviews the automated load capacity calculations and makes any final adjustments. They can then print and attach the load calculation as part of the permitting package. In many cities, this is a required and often a time-consuming process. 

The tools also bring consistency to how the local calculation is done, which could impact the safety of an installation. Puthenveetil also says this AI-based automation can perform a load calculation in minutes. It may have previously taken hours to aggregate the data, organize it, and then run it manually.

The new AI capabilities are also being exposed via API to streamline integration with OEM websites, work order management apps, and third-party field applications. Puthenveetil says:

We also expect the model to expand to the commercial electrification space. By taking the API approach, we are keeping an open mind to integrating with other systems to help remove friction in the electrification equipment installation journey. 

The dawn of virtual power plants

A big challenge for power companies lies in figuring out how to reduce the burdens of maintaining required power levels during peak demands. Virtual power plants are a kind of philosophical evolution for rethinking the role that homes and offices can play in mitigating the requirements for expensive physical wires and their associated costs. Part of this is figuring out how batteries can bank power during periods of lower demand and pay it back at peak. It’s also about automating the process of turning off power-intensive loads at peak demand. 

Schneider Electric CTO for Innovation Scott Harden explains:

By providing homeowners and contractors with a better understanding of the propensity for a home to deploy these technologies and streamlining adoption, the resource pool available for virtual power plants will continue to grow.  The US Department of Energy has set a target of 20% flexible behind-the-meter resources as a key requirement to meet our decarbonization goals.

In the US, the Federal Energy Regulatory Commission Order 2222 is moving its way through the regional independent service operator rule-making process to help include home and business infrastructure in new grid considerations. As a result, more energy service providers are looking for ways to facilitate energy aggregation using virtual power plants.  

Lessons for other industries

One essential takeaway for other enterprises is that there are plenty of opportunities for automation and improvement using traditional AI and machine learning without a need for generative AI. Most of this value comes from figuring out how to think deeply about the data and use cases. Puthenveetil's key takeaways for other industry leaders include thinking about the following:

• Domain expertise and AI efficacy: The effectiveness of AI models is greatly enhanced when combined with deep domain expertise. For CIOs and CISOs, this underscores the value of integrating AI with their existing knowledge base to drive better decision-making.
• Value of proprietary data: Utilizing proprietary data can tailor AI models to be more precise and relevant to specific corporate needs. This is crucial for large enterprises like Walmart and Boeing, where customized solutions can bring significant competitive advantages.
• Choosing the right use case for AI: The impact of AI largely depends on selecting appropriate use cases. For a CIO or CISO, this means identifying and prioritizing those areas within their organization where AI can provide the most strategic value, whether in improving operational efficiency, enhancing cybersecurity measures, or driving innovation. 

My take

The apartment where I live has a very old electrical system, which I recently discovered suffered some engineering shortcuts. In particular, the shower's electric water heater would consistently trip a breaker every time someone turned on another water outlet. Worse, it emitted an awful plastic smell every time someone took a shower. 

After a skilled electrician visited, he surmised that the previous team had installed too small a wire. A second team returned and upgraded the wire, and everything has worked just fine ever since. The point of all this is that simply running electrical wires is a relatively simple affair, but figuring out how to do so safely requires a certain degree of competence and deep insight. This will probably be even more of a problem as more people attempt to install home chargers, heat pumps, and other net-zero-friendly technologies. 

Anything that can help automate this process safely will go a long way toward helping us meet Net Zero goals - and more importantly taking showers that don’t cut out when someone flushes the toilet.