Industry
TCS Research
TCS Research: Inventing to build sustainable futures
HIGHLIGHTS
- Observability is the essential foundation needed to enable hyper operations.
- Observability workflows powered by AI and machine learning assume that data and information are continuously available from the sensors, which is not true.
- Inferences derived from partial observation could lead to incorrect decision-making, affecting enterprise operations and growth.
- Enterprises need to build intelligent observability capabilities by combining human expertise and machine perception to gain a holistic understanding of its operations.
- To achieve this, we recommend an intelligent, multimodal perception system to integrate data from multiple sources, making it easier to generate consolidated insights across complex environments.
On this page
Observability: The foundational layer for achieving hyper-operations
Technology has quietly transformed enterprise operations. In fact, traditional operations technology has evolved to what we call hyper operations, which stands for AI-powered autonomous planning, operations, maintenance, and sustenance at hyper-scale. With hyper operations, traditional IT has extended into new areas of enterprise business. Consider operations for asset inspection in the transmission network of a utility business. Going beyond mere control and monitoring of assets, a hyper operations system also acts as an operations and maintenance platform that can create situational awareness by collecting, managing, analysing, and interpreting heterogeneous data using AI.
At the heart of all this is observability, which forms the foundation for hyper operations. Observability is the ability of a system to sense, perceive, and reason all aspects of a phygital system. Essentially, it reduces efforts in detection and identification of entities or objects of interest, identifying attributes of an entity, understanding its relationship with other entities, and the behaviour of the entity. In addition, it offers insights into constraints on the object of interest - something that’s critical for modelling a system that governs the attributes and interactions.
To understand why observability matters, consider asset inspection. The goal is to detect defects and act on them. Key entities include assets like transmission lines, whose attributes, such as thickness or temperature, must be monitored. Their relationships, for instance, with tower infrastructure, and their behaviour over time (like line sag) reveal deeper system conditions. Constraints such as structural integrity further shape how these assets perform. Together, these elements form a conceptual model for systems that blend structural and dynamic components to enable reasoning, simulation, and action. Observability is what makes this knowledge possible. However, enterprises usually end up with “partial observability.”
Siloed data limits enterprise-wide insight and informed decision-making
The ability to generate relevant business insights using technology fundamentally presupposes access to timely information from various phygital systems. Observability workflows powered by AI and machine learning assume the data and information are continuously available from the sensors, which isn’t true. There are multiple reasons for this:
- Ideal sensors do not exist: Sensors don’t capture every aspect. Manifestation is often observed as secondary symptoms.
- Data scatter: Information is derived from inconsistently captured data across multiple independent systems serving various business channels.
- Multiple modalities: Data is captured in structured (logs, tables, relational databases, and time series) and unstructured (text, images, videos, documents, and emails) forms across various independent systems.
- Data loss and insufficiency: Firstly, data is not uniformly collected across locations, leading to sparse data. Secondly, unreliable communication networks and latency result in systems receiving sparse and inconsistent information.
- Scatter amplification: The internet of things has brought in big data and non-relational database systems (NOSQL) databases that amplify the data scatter problem.
More often than not, every sensor is an independent system by design. Think of sensors deployed to monitor utilities assets. A drone-captured infrared image of a transmission line creates a map of the thermal profile. Usage patterns captured by smart meters can give insights into demand. However, an efficient function requires multiple sensors and systems to come together to make optimal decisions for operations and maintenance. In most cases, there is usually a need to fuse sensor information to yield enhanced insights and derive benefits. But with each siloed system narrowly focused on generating insights specific to its target environment, the enterprise-wide macro view is lacking.
When every business unit in an enterprise pursues AI and digitisation differently, data becomes scattered, and decisions slow down. This fragmentation creates partial observability - an environment where neither AI systems nor enterprises can act with confidence.
Building situational awareness through intelligent multimodal perception
So how do we enable situational awareness and full observability? It is through intelligent multimodal perception. These capabilities include pattern discovery, prediction of future states, and root cause analysis, by combining human expertise and machine perception to understand the operations. The current use case-based approach limits the application of AI in operations technology.
An intelligent, multimodal perception system can help integrate data from multiple sources, making it easier to generate consolidated insights across complex environments. Such a system can address these challenges by:
- Integrating multiple sensors and data streams for reducing blind spots, helping mitigate information deficiency. Data and information fusion help close the information gap.
- Achieving deep insights by understanding hidden relationships among sensors and fusing multiple modalities. This removes blind spots to achieve a more comprehensive view than simply aggregating individual sensor inputs. Graph analytics and emerging multimodal graph neural networks will allow better situational understanding and root cause analysis in, say, an asset damage detection scenario.
- Automatically selecting the right pattern mining workflow enables adaptivity and avoids partial observability in a changing environment. For example, while detection of infrastructure defects can normally be achieved using a drone-mounted multimodal camera, automated dynamic workflow selection ensures the optimal multimodal analytics pipeline is chosen even as conditions shift.
- Dynamically visualisation and knowledge modelling improve human-machine interaction. While sensors often complement one another, many situations still require contextual human input. Knowledge modeling injects this context, and dynamic visualisation or conversational interfaces present derived insights in a human-friendly form.
From fragmented insights to hyper operations readiness
For enterprises looking to achieve hyper operations and greater operational excellence, embracing an intelligent multimodal perception system will be the smart thing to adopt. By effectively addressing partial observability, it will equip enterprise systems for pattern discovery at scale and help overcome the challenges of existing monitoring techniques with agent-based perception machines that proactively seek, holistically understand, and adeptly respond to dynamic environments.
When infused with multimodal perception, systems can deliver on the promise of situational awareness and give decision-makers an edge with valuable insights. Human-augmented AI ecosystems and autonomous AI systems won’t seem so far off then.