Data acquisition strategies are evolving in the era of COVID-19, driven by increased requirements for remote data visualization and real-time, data-driven decisions. Human-machine interface (HMI) and supervisory control and data acquisition (SCADA) systems are becoming even more important to achieve digital transformation because they can perform analytics on edge devices, providing the agility and resilience fundamental to success.
Connecting with Sensors New and Old
Traditional instruments and newer Internet of Things (IoT) sensors installed on field equipment are in close proximity to edge devices, which can in turn capture large amounts of produced data such as pump pressures or machine operating conditions. International Data Corporation (IDC) forecasts that 79.4 zettabytes of data will be created by the IoT alone in 20251 but all this raw data will not necessarily produce insights. Instead, value will only be created by obtaining and applying information, knowledge, and insights derived from analyzing raw data, empowering end users to improve processes.
Smart sensor solutions are available today for compressing, filtering, or converting real-time signals into the desired format for analysis; however, there are many legacy devices already in service with proprietary protocols that also need to be included for a complete data-driven transformation. Because HMI/SCADA software installations are already near the edge, it is natural and convenient to use this software for both communicating with legacy devices and collaborating with smart sensors to support real-time, data-driven decision-making.
Bringing Analytics to the Edge
HMI/SCADA software capable of performing advanced analytics on edge devices will play a pivotal role in the control, visualization, and formulation of insights and knowledge from real-time data. Because it is already used for visualization and control, HMI/SCADA software currently provides monitoring and control of operations (Figure 1).
Moving forward, HMI/SCADA software can supply essential context to realtime data in support of data-driven decision-making. It can also detect known patterns and discover anomalies, and can predict and warn operators of impending failure well in advance. Inference and local action can be handled on the edge device or edge server, while aggregated data or predictive modeling can be performed in the cloud. Advanced analytics expands the traditional capabilities of HMI/SCADA so users can understand the root cause of events and behaviors as well as predict future conditions.
Deploying mobile HMI/SCADA applications with analytics at the network edge offers organizations an additional level of flexibility to support intelligent mobile services for remote workers. These mobile applications can leverage rich data collection from remote devices and share it with an edge server for further aggregation and analysis before sending it to the cloud for machine modeling and other analyses.
Close integration of analytics with HMI/SCADA applications, defined as embedded analytics, is on the rise because of the need to transform data into knowledge. Allied Market Research forecasts the market for embedded analytics to expand from $25.13 billion in 2016 to $60.28 billion by 2023 2. When traditional HMI/SCADA software is enhanced by embedded analytics capabilities, users get a better application experience by merging insight and action into the same application.
Results of embedded analytics include key performance indicators (KPIs), statistical evaluations, and alerts close to where operators do their work and where decisions are made. This demand to push digital transformation closer to the edge where the data is collected exists because of the pressure to make data-driven decisions quicker. With the need for accelerating to near-real-time responses, operators no longer have time to return to the control room for analysis.
Machine Learning and Algorithms
Data analytics extracts meaningful insight from real-time and other data sources. Machine learning is a form of analytics using algorithms to extract data, learn from it, and then forecast the future based on the historical data. Algorithms turn a data set into a model. The optimal algorithm training — or learning — method depends on the kind of problems being solved, the computing resources available, and the nature of the data. Two leading learning methods are supervised and unsupervised.
With supervised learning, an algorithm is presented with a set of inputs along with their desired outputs (also called labels). The goal is to discover a rule that enables the computer to essentially break down the relationships and learn what input data is mapped to the outputs and how.
With unsupervised learning, an algorithm is presented with a set of inputs but no desired outputs (labels), which means the algorithm must find structure and patterns on its own as it assesses and classifies thousands of data points based on discovered patterns. There are four traditional categories of analytics: descriptive, diagnostic, predictive, and prescriptive, with a fifth emerging in the form of cognitive.
Descriptive analytics answers the question of “What is happening?” Based on real-time and past data, it garners insights on how the process is performing by providing context to the data. Based on real-time data, HMI/SCADA applications using descriptive analytics provide visualization of what is happening, annunciate alarms for the operators, and deliver details such as the date/time of occurrence, associated values, and machine information.
Diagnostic analytics builds on descriptive analytics to answer the question, “Why did this happen?” Diagnostic analytics uses statistics to find patterns and offer insights into real-time data. Typical uses are identification of anomalies and root causes (Figure 2).
Predictive analytics builds on diagnostic analytics to answer the question, “What will happen in the future and why?” Predictive analytics leverages the same historical data as the prior two types of analytics to build mathematical models that can be used to make inferences about what will happen in a future, forewarning operators of future events impacting productivity (Figure 3). There are several tools used for predictive analytics including:
Anomaly detection, which answers the question, “Is the behavior anomalous?” It finds data that does not conform to an expected pattern.
Classification, which answers the question, “Will this machine or process fail?” It classifies data into binary categories such as “Does this product have a defect? Yes or No.”
Multi-class classification tries to answer the question, “Will this machine or process fail for reason X?” It classifies data into one of three or more categories such as “What kind of quality defect does this product have? Minor, major, or critical?”
Regression answers the question, “How long before this machine or process fails?” Regression uses statistical processes for estimating the relationships between a dependent variable and one or more independent variables.
Prescriptive analytics builds on predictive analytics to answer the question, “What should we do?” Prescriptive analytics provides models to inform operators regarding recommended actions. Optimization and simulation algorithms are often used for prescriptive analytics. HMI/SCADA working with prescriptive analytics can prescribe several different possible actions and guide operators towards a solution.
Cognitive analytics builds on prescriptive analytics to answer the question, “Why should I do it?” Cognitive analytics uses techniques of self-learning algorithms and deep learning to emulate human thinking.
Descriptive and diagnostic analytics use past data to explain what happened and why it happened while predictive, prescriptive, and cognitive analytics use historical data to forecast what will happen in the future along with what actions should be taken to affect a specific outcome. In many cases, multiple analytics and algorithms are used simultaneously, with results aggregated for better decision-making.
HMI/SCADA software deployed on an edge device can be a key factor in providing analytics to drive organizations toward better data-driven decision-making even from remote locations, improving competitiveness.
If there is a silver lining to the COVID-19 pandemic, it is the ways companies are learning to adjust business models to more real-time, data-driven decision-making. By creating more remote visibility into processes utilizing people, equipment, raw materials, and facilities, companies are finding they can run more optimally by providing closer ties to customers, employees, and suppliers.
This article was written by Bruno Armond Crepaldi, chief technology officer at ADISRA (Austin, TX). For more information, visit here .
- June 18, 2019 - The Growth in Connected IoT Devices Is Expected to Generate 79.4ZB of Data in 2025, According to a New IDC Forecast.
- June 1, 2020 - Global Embedded Analytics Market to Reach $60.28 Billion by 2023: AMR.