Consensus-based Sensor Infrastructure for Safety & Data accuracy

Internet of Things (IoT) has paved the way for capturing data from numerous devices. Intelligent, connection ready hardware is now embedded in devices to transmit data to external consumers.

The devices which were restricted by the type of data it can send, type of destination it could send to, the medium it must use to send data are the things of the past. Today, devices expose various interfaces over the network for data capture. Many of these devices are certified to support industry standards for communication which makes it easier to integrate with a multitude of consumers.

While this is great, if the care is not taken to come up with a topology that ensures data integrity, the data collected might reflect false insights. Decisions based on false data might be fatal especially in healthcare, transportation, financial & other mission-critical domains.

What is data integrity? In simple terms, ensuring that the data is accurate, consistent & reliable. What if the data is inaccurate? It leads to unexpected behavior. It is of no use. It can cause havoc if not detected & filtered at the time of ingestion. A recent example being the Boing’s 737 MAX aircraft crash. Investigators found that relying on a single sensor & the faulty data from the sensor caused the MCAS (Maneuvering Characteristics Augmentation System) to misfire, which could be one of the reasons for the recent crash. One obvious way to handle this is by adding an additional sensor. Does this solve the problem? Let's take a use case (Topology 1):

Road safety using a visibility sensor: 

The visibility sensor measures the distance up to which an object can be clearly identified. A quick explanation can be found on the UK MetOffice website here.  

For our use case, let's consider measurements:

Sensor 1: Visibility distance: 500 meters

Sensor 2: Visibility distance: 200 meters

Which value would you trust and communicate? Yes, today’s sensors are mostly accurate. What about Sensor tampering? The tampering could be with the hardware or software. What if one of these sensors is tampered to transmit false values? The false readings could be life-threatening. In winter months, accidents due to fog and smoke are common across the US and many other countries. 

To avoid the above scenario, I propose adding a third sensor to the mix. How does that help? Let's have a look at Topology 2: 

Let's consider the following measurements:

Sensor 1: Visibility distance: 500 meters

Sensor 2: Visibility distance: 200 meters  

Sensor 3: Visibility distance: 195 meters

Looking at the above measurements, we can infer that the visibility distance is most likely about 200 meters given 2 out of 3 sensors concur with it.

Programmatically, we can achieve this using a consensus algorithm. The idea of consensus is to achieve an agreement from the higher number of nodes conforming to the outcome. By deploying at least 3 sensors, we achieve a level of consensus. This is the same mechanism that is used in blockchain implementation. This logic fits well on an Edge compute to act upon the data in real-time.

There are numerous advantages of the proposed topology including:

1. Real-time insights based on accurate data
2. Data consistency over time
3. Detecting tampered sensors
4. Detecting hacked software stack
5. Locating the faulty sensor
6. Measuring sensor life expectancy

Yes, in the case of Industrial IOT, where thousands of precision sensors are used, an additional sensor would exponentially increase the cost of deployment, and compute requirements to implement consensus algorithm. While this is true, the compromised sensors and inaccurate data will lead to irreparable damage.

The proposed consensus-based sensor topology can be selectively deployed at critical points to manage costs and can be gradually scaled.

I would love to hear your thoughts. Any type of feedback would be greatly appreciated.