Much of the current thinking for making systems “smart” takes advantage of inexpensive hardware and fast wireless networks in a loose design approach that gives little thought to the actual problems facing the end user. In energy monitoring, for example, utilities have installed millions of smart meters that provide little or no actionable information to the facilities owner to help manage or reduce consumption. Engine rooms of modern naval vessels bristle with sensors, each generating data that is faithfully recorded, and typically ignored. Data meant to help becomes a hindrance when operators have to debug hundreds of sensors to find the source of a false alarm. From the smart home to the smart grid, sensors generate clouds of data that overwhelm instead of inform. Databases storing this information have grown so large that analyzing them has become an academic discipline in itself. Even sophisticated players in the data analysis market find themselves unable to capitalize on the promise of cyber-physical systems with both Microsoft and Google cancelling their respective energy monitoring projects soon after inception. Designing truly functional cyber-physical systems requires both an analytic mastery and practical expertise. Industry fails to appreciate the complexities of sensor design looking for quick profit with off the shelf components while academics focus on isolated algorithms and circuits at the expense of the system. Our research combines academic rigor with a mastery of the practical details required to deploy a complete system from the front-end sensor and signal processing to the backend network encryption and server architecture necessary to bring actionable information to the end user.
System design begins with the sensor. Real time electricity meters enable energy conservation but require current and voltage sensors that are expensive and inconvenient. These sensors use Ohmic contact to measure voltage and geometric isolation of each phase to measure current. Installing such a system involves a trained electrician and a service interruption that often costs more than the savings gained. We are designing non-contact sensors that can measure both current and voltage from outside the insulation of a power line making it safe and easy to install. Precision electromagnetic sensors measure dipoles escaping the power line to reconstruct the current and voltages on each phase.
Water flow meters are similarly intrusive, requiring a plumber to insert monitoring equipment inline with the pipe. Our retrofit sleeve converts a standard meter used for utility billing into a high bandwidth flow rate monitor. The system attaches with a zip tie and matches the accuracy of industry standard inline devices.
Both the energy meter and flow rate sensors are designed to be accurate as well as non-intrusive. As hardware costs continue to decrease, sensor installation and maintenance begin to dominate the total cost of ownership, making non-intrusive designs particularly advantageous. Non-intrusive systems also provide exciting possibilities for in-situ diagnostics where physical and operational constraints make traditional sensor platforms impractical. We are working closely with the US Navy and Coast Guard to develop better diagnostics for marine motors and generators. The current practice requires intrusive instrumentation that can only be done in port. We are designing sensors that can provide similar diagnostics without electrical or mechanical connection to the equipment allowing the crew to retrieve diagnostics while underway. This gives operators real time assessment of machinery health in forward deployed environments where such information is critical for survivability.
Embedded Signal Processing
Non-intrusive also means reducing the total sensor count. By applying the appropriate signal processing, a single sensor placed in the right location can provide data equivalent to dozens of distributed sensors that would require complex communication protocols, power, and of course installation. When only one sensor is required, more resources can be devoted to its design. Recent advancements in mobile computing have lowered the cost and power consumption of microcontrollers to the point where powerful 32 bit systems with floating point DSP can be directly embedded in the sensor platform. The next generation of sensors should not just measure and transmit; they should process their own data locally. This design enables much richer signal acquisition because data does not have to move across a network. It also eliminates the ethical complexities of using external storage providers like Microsoft and Google, who may have ulterior motives with user data. Moving to embedded environments requires efficient database and computational frameworks. Our research focuses on new signal processing techniques for these resource constrained environments. Most smart power meters sample current and voltage at 1Hz or slower. Our power monitor prototype shown below samples current and voltage at 3 kHz, fast enough to identify load transients and perform equipment diagnostics. All of this computation is performed locally on the device with results stored in a custom database designed for high bandwidth time series.
Distributed Cloud Architecture
Without a means to communicate, non-intrusive sensors, however sophisticated, do not provide a practical solution. The third component of our research is the design of a new type of cloud architecture that connects end users with their remote sensors. In typical usage “cloud” describes a central server hosting content that is consumed by remote clients. When the content is located on the sensors themselves, the “cloud” acts as an intermediary instead of a repository. We envision sensors that communicate directly with end users employing encryption to verify the confidentiality and integrity of the data path. This is in stark contrast to current cloud models where users have little to no control over where or how their data is used. Our cloud manager, shown below, provides a presentation interface that clients access through their web browser. This interface is then dynamically populated with data retrieved from the user's sensors.
To deliver truly actionable information, sensor platforms must be customizable. Not only are the requirements of residential, commercial and industrial consumers quite diverse, they also change quickly. In centralized frameworks users are at the mercy of the service provider for data processing and analytics. By moving the data and processing tools out of the cloud and onto the sensor, the user is in full control of the data path. To realize the benefits of this new design approach we are developing a suite of tools to support user-designed signal processing and data visualization. This work has two primary components. The first is an application programming interface (API) which allows users to inject custom code or “apps” into the sensor’s data processing pipeline. The second component is a web-based integrated development environment (IDE) where users can write, debug and share sensor apps enabling a new type of decentralized interaction where data is private but the code is collaborative.