Leveraging the Data from IoT
By Mike Shemancik, CIO, ABT Power Management
Mike Shemancik, CIO, ABT Power Management
There is a lot of hype and excitement about the “Internet of Things” and the promise it brings around collecting and triangulating data from whatever array of sensors and devices are applicable to your business. The technical and logistical barriers which made the collection and management of such data cost prohibitive in the past have been rapidly falling by the wayside. Even a smaller company such as ours with less than 100 employees can now implement and manage a substantial IoT infrastructure which would have cost 5-10 times as much not long ago.
At ABT Power Management, even though we are small, we’ve always been a pioneer in our industry and we’ve been collecting sensor data about the assets we maintain and analyzing it for over a decade, long before this practice was known as IoT. These assets are industrial batteries and charging systems used to power forklift trucks and other heavy material handling equipment in warehouses and large distribution facilities. Our flagship program which we call GuaranteedPower ensures that our customer’s power needs are met in a manner that enables smoother operations and lowers their total cost of ownership. However, the success of GuaranteedPower significantly depends on closely monitoring the performance of these assets. Fortunately, our founder and CEO, Ken Fearn, had the foresight to recognize the value of this sensor data long before anyone else in our industry did. However, it’s only been in the past two years that it’s been possible for us to build a truly scalable infrastructure and more efficiently leverage and operationalize the data. Our IoT solution is transforming ABT’s business in multiple ways, including increased agility, decreased labor costs, and enhanced scalability.
If you’re considering embarking on an IoT initiative, you may be questioning how you should begin. You could consider the following approach:
1. Develop a business case with a compelling ROI. Think about which questions, if you could answer them through sensor data, would improve customer satisfaction, improve operational efficiency, enable increased scalability, and/or decrease costs?
Obviously, in most organizations, you’re not going to get the support or funding you need without presenting a compelling ROI. Think about an application of the data that can deliver measurable results and use that as a beachhead for demonstrating value and garnering support for future expansion.
2. Analyze the availability and quality of data. Which data sources would be relevant? Is it really practical and cost effective to ingest the data? What are the gaps in the data? How much transformation is necessary to get it into a usable form? What existing data sources can be correlated with newly acquired sensor data?
Although great advances have been made in a short period of time with respect to collecting device data, it is still far from clean or easy in many industrial environments. For example, in our case, we’re monitoring industrial lead acid batteries that are used to power forklift trucks and these batteries typically operate at between 110 - 135 degrees Fahrenheit and contain highly corrosive acid; not exactly an environment conducive to monitoring devices.
As one of our engineers recently quipped, “We’re just trying to put electronics on a hot bucket of acid. What could possibly go wrong?” The consequence is that our monitoring devices had a high failure rate and consequently, data packets were frequently lost. We’ve had to look at sourcing more rugged devices and in the meantime, factor that issue into our analyses and inform data consumers about the level of completeness of the data when they are making decisions (a “data quality meter”, if you will). The lesson is that you need to thoroughly scrutinize the quality of your data and be realistic about the limitations and how you’re going to handle them.
3. Determine the appropriate method for leveraging the data. You can collect mountains of data but if it isn’t actionable, it provides no real value. There are a number of methods available for acting upon the data and based on the business case you put together, you’ll need to determine the one most appropriate for your business. You might consider the following scenarios and whether they apply to your situation:
Scenario #1: You have internal resources who possess expert knowledge about the devices you’re monitoring, but you do not have a scalable means of deploying that knowledge or reacting to changing needs in a timely manner.
Solution: Deploy a Decision Engine – In our case; we had engineers in the company who had acquired decades of expertise about how batteries perform and how they should be maintained. However, the problem we faced was that as the company’s asset base rapidly grew, manually scanning growing volumes of data was no longer a scalable solution for applying their expertise. Our approach was to codify their knowledge in a set of rules and implement them in a decision engine product, an approach which historically has been largely consigned to the Financial Services and Healthcare industries.
By centralizing this logic, ABT can easily change policies in one place if needed and deploy them instantaneously and consistently across all customers and assets. Furthermore, rules can be maintained by the engineering staff themselves without requesting IT development resources, resulting in increased speed and agility to react to changing needs and opportunities.
Scenario #2: Somewhat the opposite of Scenario #1, you have lots of (presumably quality) data available, but have limited internal expertise to analyze it.
Solution: Employ Machine Learning – Various machine learning methods such as Decision Trees and Regression Analysis can analyze your data and detect patterns not discernible to human analysis. If you don’t know exactly what insight your data will provide or, if in the case like ours where we are using it as a supplement to our human-based expertise to see what additional insights we can glean, this may be an appropriate tool for you. However, machine learning is not a magic solution where the computer does all of the work for you. There is still a great deal of human effort involved in “shaping and training” the data (i.e., using subject matter expertise to get it into a form conducive to analysis and running good control data through it to refine the model).
Scenario #3: You’re performing analysis of the data, but it’s not producing any tangible results for your business.
Solution: Operationalize the Data by integrating the data with your transactional systems. In our case, we are integrating our IoT monitoring solution with our ERP system so that service work orders are automatically spawned to dispatch maintenance staff when certain conditions are detected in the assets. Systems integration efforts are frequently challenging and we’re in the process of evaluating IPaaS alternatives to determine the best means of effecting the movement of data between systems. Think about which of your operational systems can be positively impacted by your sensor data.
There are a number of other options for acting upon your sensor data, including use of a visual analytics tool (we are using Tableau) for advanced ad hoc analysis or subscription-based reports for stakeholders who need a consistent, repeatable snapshot of the data important to them, like our service technicians who need to see a daily report of the performance of the assets at their particular site.
"Implement and manage a substantial IoT infrastructure which would have cost 5-10 times as much not long ago"
Bottom line, you’ll need to develop a compelling business case and then determine which method for handling the data is the most appropriate for deriving organizational benefit.