Hello, good morning. Good afternoon. Good evening everyone Depending on where you are joining from. my name is Akshat Kapoor and I welcome you to my talk on real time analytics in IIoT, before we dive in a bit about myself, I have been in the tech industry for over 20 years now, working with some great organizations like Alcatel Lucent Enterprise, Dell Labs, and Nokia, my career has mostly focused on product management. guiding product portfolios and finding innovative solutions that drive growth for the company at Alcatel Lucent. I am responsible for the full range of, Ethernet switching products from campus land and industrial Ethernet to data center. Switches. My current areas of focus are AI, ML, and industrial IoT. And that relates to the topic for today. And that brings me back to the topic. IoT and industrial IoT are transforming industries by connecting devices and creating smarter and more efficient operations. Today we will uncover its potential and go through the journey of how real time analytics is revolutionizing the industrial processes. We will start with some building blocks of the architecture that enables real time analytics and go on to see how all these pieces put together enable the real time decision making for predictive preventive maintenance. for optimizing operations and saving money for various industries. So let's talk about IoT or industrial IoT in modern industries. The internet of things or IoT has evolved as a cornerstone of industry 4. 0, which is connecting physical devices to enable smarter operations. The convergence of IT and OT systems drives automation and efficiency, offering key benefits like predictive maintenance, operational optimization, and real time decision making. Examples of its transformative effect can be seen in industries like manufacturing, logistics, oil and gas, utilities such as electrical utilities, and smart cities. Real time data processing and why it matters. Real time data is critical in dynamic environments. where timely insights lead to competitive advantages. However, businesses face challenges like scalability, interoperability, and handling diverse data coming from the sensors. By overcoming these hurdles, organizations can unlock faster decision making and operational success. A manufacturer might use sensors throughout an assembly line to predict when a machine will likely break down. Or an oil and gas organization might use predictive analytics on data from sensors installed across a pipeline to help identify corrosion and damage. Industrial IoT brings together connected intelligent devices and analytics in a way that allows organizations to monitor Collect, exchange, analyze, and deliver valuable new insights about their processes and systems. Now, we'll talk about some of the benefits of IIoT real time analytics, and how we are utilizing zettabytes of IoT data. Now, as we know that the IoT devices produces a ton of data, And, data from these telemetrics about a device or machine's condition, when combined with historical performance records, can offer deep insights that predict and help prevent failures. For instance, a common method to prevent equipment downtime is replacing a component as it approaches its mean time to failure, or MTTF. By leveraging machine learning based analytics on real time IoT data combined with product history information, it is possible to detect early warning signs of potential issues in parts before they reach their failure age or MTTF. More efficient and automated QA processes reduce cost while improving product quality. IoT data can help identify inefficiencies. that when resolved lead to increased output. Now let's look at the basic building blocks of some of these IoT architectures. You can call it a three layer architecture that you see in the blue on the left hand side, or you can call them four if you want to follow the architecture on the right in green. That includes the application layer that includes the applications and all that run on top of analytics and the management layer. But the basic three layers are the same in both of them. So let's start with the sensing layer. The sensing layer is the one which collects data using edge devices and sensors. Firstly, there is the issue of data preparation to harness the value of IoT and deliver innovation driving insights. Organizations must quickly prepare and standardize great volumes of disparate unstructured data. This is a common problem across all any AI ML based system, which is required for analytics. But it is more compounded in industrial IoT because the sensors from different parts of the manufacturing line or wherever they are installed have very disparate data. It's, some have geolocation tagging enabled, some have not, and so on and so forth. As we all know, the data preparation takes up to 80 percent of the time and resources in a data analysis operation. In heterogeneous IoT systems. And devices exhibit a wide variety of characteristics and differ in their computing, communication, and storage capabilities. Now let's talk about the communication layer. The communication layer ensures efficient data transmission through robust protocols. The communication layer, supported by multi service gateways, forms the central network for IoT data transmission. Managing both uplink and downlink of feedback from the, from the analytics layer due to the diversity of the IOT devices, various communication protocols are utilized within IOT systems. And I'll be talking about this in another slide further down and finally, the the analytics layer processes this data that is coming from the sensing layer to the communication layer and using edge cloud or hybrid methods, balancing latency and performance. And we will also be talking about this further in, in this presentation. Real time analytics in IoT systems emphasizes designing system architectures capable of performing data analysis and delivering responses within a specified timeframe, referred to as the deadline. Depending on this application, this deadline can vary, ranging from nanoseconds in computer network communications, To milliseconds in medical diagnosis real time industrial iot systems are of two types predictive and prescriptive. And they are different from diagnostic or descriptive kind of analytics because they have to work in real time. Traditionally, there are two types of architectures used in the analytics layer, which is the edge, cloud hybrid or the cloud. There's also a third type of, new kind of, analytics architecture, which is based on simulation and that's the most upcoming and promising technology that is there today and I will talk about it in another slide. So let's talk about the first layer which is the sensors from where the actual data is collected. The specific type of sensors that are deployed within an IIoT system depends on the machinery being monitored and the parameter most critical for predicting potential failures. So common type of sensors that are employed are vibration sensors, temperature sensors, pressure sensors, or current and power consumption sensors, or any, which is required for the type of environment other than the type choosing the right type of sensor, the strategic placement of the sensors throughout the machinery is also essential for capturing the most informative data. Now we talk about the second layer, which is the communication layer. which is required for real time analytics. Automating processes by integrating the operational technology OT with information technology IT as mandated by industry 4. 0 brings stringent real time demands for industrial networks making predictable and reliable packet routing very essential. Traditional IP networks which operate on a best effort basis are insufficient for these real time needs. and require specialized protocols and standards to meet these performance expectations. Depending on the type of industrial network, there are different type of requirements for the network to perform. For example, autonomous systems, such as robots and drones, require high predictability and medium to high latency tolerance. Remote control applications need low bandwidth, but highly reliable networks. Data analysis and monitoring require high bandwidth for sensor data aggregation. Traditional embedded systems only have limited computing capabilities due to power and space constraints. To this end, currently, especially in the field of mobile edge computing, considers the offloading of delay sensitive tasks to local servers. Worker safety systems emphasize latency and reliability to react promptly to hazards. Now talking about security, which is true for any kind of network, but more so in on industrial networks, not only the data on the network should be encrypted, it should be able to detect network intrusions in time. And last but not the least, the, as networks are used to converge IT and OT systems to build a distributed system, everyone should have a common and shared understanding of time and real time for real time networks to work. No one single system today is capable of meeting all the requirements. And, of course, the type of network that is chosen is based on, the type of end system and what its requirements are in terms of predictability, latency, bandwidth, so on and so forth. as I said, real time analytics requires real time networks. The current package switch network, which is based on best effort, is not sufficient to meet the requirements of real time analytics in IIoT. Various approaches can be applied to improve real time IIoT networking, and there is no one size fits all here and depends on the end user environment. Some of these approaches are software defined networking or SDN. As industrial networks grow more complex with mixed computing systems and real time demands, SDN offers a solution for managing this heterogeneity. Next comes the TSN and this has to do with the shared sense of time and a common and more secure way of for networks to be able to deliver the time across all the devices whether they are IoT devices or IT systems connected on the network. Time sensitive networking or TSN is a set of IEEE standards for Ethernet enabling networks to achieve bounded latency and minimal packet loss for critical traffic at the data link layer. It ensures lower jitter and more orderly packet delivery. Additionally, it incorporates precise and detailed quality of service mechanisms. New standards have evolved that have improved some of the shortcomings and the lack of flexibility of TSN by integrating it with other systems such as OPC or adding profanate support, which is a mechanism for visibility. of IT and OT systems in a common network or in a common management plane. OPC or Open Platform Communications Unified Architecture is an interoperability standard for reliable information exchange. And has gained a lot of momentum for discussion on IIoT networks now talking about wireless technologies. So a very natural step is to, transition this TSN methods or time sensitive network. networking methods to use to be working with wireless technologies. similarly, the I iot devices, which require 5G wireless connectivity. The network slicing there allows for logically separated virtual networks over the same physical 5G network. And last but not the least, there are some technologies that have evolved on the device layer, such as. MAC layer filtering that mitigates the effects of best effort traffic reception on real time tasks. So you see the technologies on the networking side have to evolve to manage the data coming from the sensor to enable the real time analytics. And as I said, there is no one size fits all, and it all depends on the end user environment to adopt the right type of network and the right type of technology to enable this. Okay, now we talk about the analytics layer and what are the different type of architectures that are used in enabling the analytics The first one is the cloud based architecture. The cloud based architecture is a centralized, processing in remote cloud servers, provides high computational power, but there is a challenge here, which is there is a high latency, there is a limitation of bandwidth, and the connection could be unstable, which may not be suitable for a real time application. Next comes the edge cloud collaboration or the edge cloud collaborative architecture. This combines local edge computing with cloud resources. Edge devices process data closer to IoT devices. which reduces latency and bandwidth requirements. Cloud platforms handle the complex analytics and storage tasks, whereas the edge devices manage the latency and bandwidth requirements by being closer to the IoT devices. This hybrid model optimizes task allocation for real time requirements while mitigating the challenges that are faced in purely cloud based architectures. these are the two which are, have been traditionally been used. And there is a new architecture that is emerging today is the digital twin. The digital twins are virtual replicas of physical systems, which are continuously updated with real time data to mirror the actual state of the system. At that point, they are very useful in predictive analysis, process optimization, and data driven decision making. These are combined with simulation models. for decision support systems. There are mainly two types of models used for simulation in industries. The first one is the discrete event simulation. This models allows for analyzing complex manufacturing systems by creating a model and replicating their behavior over a period of time and then producing models, producing predictions over that. The second one is a hybrid simulation method. This involves integrating the virtual model with the real world system. So imagine a digital twin replica running, integrated into a physical system as another node. And it is continuously monitoring and provides adaptive control of the system in real time. So these, this is some of, this is one of the new technologies that are emerging on the horizon and which is very useful for Real time analytics in IOT, IIOT. Now we have, gone through all the three layers. we have the data coming from the sensors. We have talked about what kind of networks are required. To send this data across with minimal latency, right amount of bandwidth and, carrying the time, and being secure. And then the third approach we saw how we, the architecture of where the analytics happens. Now we want to talk about is. What AI techniques we can use to generate the real time analytics. So there are two approaches here, which, are machine learning based approaches and the other one being the deep learning based approach. The choice of machine learning, learning algorithms used in real time analytic systems depend on several factors such as, the type of equipment being monitored, what is the intended functionality. For example, you want to do anomaly detection, or you want to do classification, or you want to predict the remaining useful life of an equipment, and so on and so forth. And also, this is influenced by the attributes of the available data set, which may not be the same for different kinds of situations. So in the machine learning approach, we have broadly two methods, supervised learning and unsupervised learning. So this is common for any AI ML based system. I'm just going to highlight here what is mostly used in these terms for industrial IOT systems. So supervised learning algorithms, as we know, Excel at pattern recognition and classification tasks for label data set. where each data point is associated with a predefined category or outcome. Support vector machines or SVMs and the decision trees, they are very powerful supervised learning algorithms that are well suited at classification tasks. and that can help identify potential equipment issues. Okay. Now, contrary to supervised learning, unsupervised learning algorithms work with unlabeled data where data points do not have any predefined categories. They are particularly effective at uncovering patterns and relationships within the data, making them well suited for tasks like anomaly detection and so on. Within unsupervised learning, techniques like k means clustering, which is a popular unsupervised learning method, is used for data segmentation and the other one being the principal component analysis, PCA. is a dimensionality reduction method that extracts the most important features from a data set. Supervised and unsupervised learning techniques provide a robust set of tools for detecting patterns and anomalies in sensor data gathered by industrial IoT systems. Now, utilizing these machine learning algorithms, systems can derive meaningful insights from real time data streams, supporting proactive maintenance strategies, and improving overall equipment efficiency in industrial operations. But deep learning that utilizes artificial neural networks with multiple hidden layers is capable of learning complex patterns and relationships within data, which often outperform traditional machine learning algorithms. in certain applications. So it all depends on your end application, whether you want to apply the machine learning approach or the deep learning technique. Of course, deep learning techniques are more complex to deploy, but they do offer several advantages over machine learning techniques such as. automatic feature extraction, improved pattern recognition, real time anomaly detection, scalability, and adaptability. Some of the architectures, which I'm going to talk about, is common with any other deep learning techniques deployed in other machine learning algorithms. But here in industrial IoT, the most common ones used are convolutional neural networks, Recurrent neural networks and of course the hybrid of the two. The advantage of convolutional neural networks is to automatically learn and extract key features from raw sensor data. Whereas the RNNs, Recurrent Neural Networks, have the ability to learn long term dependencies within the data. Within RNNs, Long Short Term Memory, or LSTM, networks analyze historical sensor data combined with equipment failure timestamps, to learn machinery degradation patterns over time. This ability enables them to accurately predict the remaining useful life of equipment. A third model, which is very promising and a new one, is focusing on integrating the strengths of CNNs and LSTMs into a unified deep learning architecture. CNNs can be utilized for extracting features from sensor data, while LSTMs leverage these features to capture temporal dependencies and predict equipment health or remaining useful life of the equipment. This hybrid approach has the potential to improve the accuracy and efficiency of AI driven solutions. So we have talked about all the three layers and we have talked about what sort of analytics to apply to get the real time digital analytics. Now is the time to see the IoT applications in action and how they are being used across multiple industries. They span, of course, multiple industries, smart systems, revolutionize healthcare, energy grids, and environmental monitoring, real time analytics, monitor traffic in smart cities, manage energy usage, and detect anomalies. Examples include predicting traffic congestion using sensor and social media data. In smart healthcare, IoT devices support chronic disease monitoring and emergency detection. The edge cloud collaboration architecture in analytics enables timely responses as seen in systems for diseases such as obstructive sleep apnea and dementia care. In smart electricity grids, Real time data ensures efficient energy distribution, and minimizing the outages. Analytics optimize demand response system for better energy management and in industrial IoT, they facilitate automation, predictive maintenance, and process optimization using low latency network architectures to meet critical operational demands. And the third one is on the digital twins. we talked about it earlier. They are helpful in optimizing production and supply chain. And they can be seen in aviation industry's use of predictive analytics to reduce the downtime. Of course, no system is perfect and there are, of course, challenges in IoT data processing. Managing IoT data involves addressing its volume, velocity, and variety of different data sets. Networking demands such as time sensitive networking and advancements like 5G are very critical to ensure seamless connectivity from a network standpoint. Additionally, there is currently a lack of standardization across platforms. which are hindering the seamless integration and interoperability of IOT or industrial IOT systems. So you could have wireless and your IOT devices could be connected through wireless as well as through wired networks. And there is no standardization across these two types of networks, for example, to do time sensitive networking. for listening. or to provide quality of service and so on and so forth. Now we understand security is paramount and so is the case for industrial IoT systems. The growing threat to connected devices requires a very robust encryption and authentication protocols to safeguard the sensitive data that is carried through the network from the sensors. Also, regulatory compliance with standards like GDPR and ISO 27001 would require would foster the trust and accountability of these systems. I'll talk about some of the future trends in IOT. So first is on from shaped by emerging technologies like 5g, digital twins, and quantum computing. On the second. Aspect is the enhanced edge cloud collaboration, which is required for optimizing task allocation and requires communication protocols to balance computational load and minimize delays. The third ones are about the advancements in privacy and security by protecting sensitive data and ensuring robust authentication mechanisms, especially in dynamic and heterogeneous IoT environments, such as leveraging blockchain for decentralized security, for example. This all will drive further innovation, making IoT systems smarter and more resilient. In conclusion, real time IoT analytics are revolutionizing industries, offering actionable insights, and have a transformative potential. Collaboration between stakeholders and continuous innovation are essential for unlocking the full potential of IoT systems. The future lies in adaptive technologies and seamless integration. Thank you all for your attention today. I hope you have a great day and you enjoy the rest of your conference. Thank you.