Deep learning facility testbed

Fast Facts


Dell EMC, SAS Institute, Toshiba


Buildings and Facilities; Energy and Utilities


Energy efficiency is an important requirement and an indicator of how energy is utilized in commercial buildings. Next generation buildings are designed to carefully optimize energy consumption by striving to achieve a delicate balance between demand and supply, with the goal of reducing energy consumption. IoT is an important enabler for realizing energy efficiency in building infrastructure. However, building installations are complex and may span several acres of land. This requires a wide variety and large number of sensors that span the entire building installation so that they can sense the ambient conditions, traffic flow, and occupancy. In addition, a building installation consists of numerous assets that consume energy, such as HVACs, fans, lights, and elevators, that are necessary for smooth functioning of the building. Deriving actionable insights from the numerous sensors and monitored assets with the goal of achieving energy efficiency can be a formidable task. Recent advances in computational capabilities that can work with these enormous datasets and derive actionable insights are making this a possibility, and creates the primary focus of this testbed.


The goal of the Deep Learning Facilities testbed is the realization of a next generation smart facility solution using Deep Learning through Neural Networks, with meaningful gains in energy efficiency, asset utilization and maintenance. Specifically, the testbed will optimize diagnosis, maintenance, and repair of monitored assets; increase energy efficiency by adjusting power-consuming services, and improve visitor experience relative to wait times and ambient climate control. Additionally, the testbed will identify optimal deep learning techniques and best practices that are both computationally feasible and efficient for multiple classes of smart facilities (individual buildings, smart campuses, smart factories, etc.). Furthermore, the testbed will provide an environment where the IIC community can leverage the testbed for their own smart facility use cases.


The deployment of Deep Learning technology poses a challenge from both a software and hardware point of view. Not only does it require very large amounts of data from multiple sources, but also domain expertise, data science, and neural network knowledge. This testbed will deploy advanced hardware and software specifically configured for Deep Learning. More specifically, Deep Learning is one of the most important technologies for the industrial IoT community. In this testbed, Deep Learning is used to train the neural network model with very large data sets from multiple sources (often in batch mode). The trained neural network creates user value with real-time analytics on sensor data.


The benefits of Industrial IoT with Deep Learning are prevalent in several key verticals today and industrial areas such as the Automotive Industry, Financial services, Industrial Cyber security. However, no commercially viable applications of ML/DL/AI technologies have been applied to the Smart Facilities space prior to this testbed. This is a first in the marketplace. Cross-correlation of multiple critical systems and assets across several deployment sites enables new insights and inferences to be drawn in an automated fashion, which is critical in environments with tens of thousands of sensors and billions of data points. The algorithmic knowledge and innovations discovered through the implementations of this testbed are intended to be applied commercially by the participants in this testbed. These net new solutions will represent tangible commercial value for multiple entities based on the foundational learning from the testbed.

The Testbed


There are a few challenges to consider from a market point of view. The deployment of such a system will require a specific wiring of the facilities. That may be expensive in some cases and may not be easy especially when we have to retrofit existing (brownfield) infrastructure. Data collection represents another challenge, given the complexity of facilities, the large number of data sources that need to be considered, and the multiple parties (HVAC Providers, Building Management System companies, Electrical and Utility Providers, Waste Management Companies, etc.) involved.


Given the expected fast growth in the number of IoT devices over the next few years (20-50 billion smart devices by 2020, depending on which analyst report you look at), we will see challenges related to the management of such edge devices, as well as reliability, security, and related risks. We also see the need for a very efficient storage, processing and management of the data that will be generated (estimated in zettabytes). There are a few technical challenges that we envision. Deep Learning is a new technology and, as such, it will require new methodologies for testing and validation with real-world use-cases. Deep learning requires important computing resources in order to calculate a training model. Optimization algorithms have not been perfected. This process will take time and experimentation. With phase 1 and phase 2, we hope to learn enough to validate the approach and its applicability to more than just one facility type. We expect to see multiple data sources as the result of the very large amount of IoT devices in operation. IoT devices will need to be autonomous for peak efficiency and will need to be reliable. The application of Deep Learning to Smart Facilities is an exciting journey, and we’d be happy to share our learnings along the way, and discuss ways to collaborate with others:

Interested in learning more about the Deep Learning Facilities Testbed? Email us!


Resource Hub