The industrial world is witnessing an unprecedented explosion in sensor data volumes, velocity and variety. The chemical industry, oil and gas industry, manufacturing industry among many others are piling on production data from hundreds and thousands of sensors installed on their assets and processes. Adding to this data deluge, machine manufacturers are embedding sensors into their products in order to collect real-time data for fault diagnosis and prediction. It is anticipated that these enormous volumes of sensor data can be harnessed in order to gain actionable insights. Reliable actionable insights can be of great value to the organization, and some of its grand promises are to
- Improve production efficiency by at least 20%
- Decrease downtime by 50%
- Reduce maintenance costs by 20%
Despite the promises, less than 5% of the industrial organizations have mature data-driven practices for harnessing data for reliable insights. Many organizations find traditional analytic tools like spreadsheets to be inadequate for taming the complex data at hand and for deriving reliable insights. In this light, it is important to revisit the fundamentals in order to realize digitalization and to build reliable and scalable AI applications for industrial applications. To this end, it is critical to ask the following.
- Do the organizations and their experts have the right tools to explore these opportunities?
- Do the organizations and their experts have the right tools to understand the quality of their data and the information in it?
- Are their experts able to process this data with their valuable domain-knowledge in a scalable way?
For sure, organizations and their subject-matter experts need to proactively discover and exploit opportunities in this data. But the reality is that the industry is drowning in data; dark data to be precise. The subject-matter experts who are expected to tame this data do not have the right tools to understand what information has been captured in the data. Most of this untamed data will not be used and will never be thrown out. Furthermore, the number of assets, processes and machines is disproportionate with respect to the number of these experts. For those working with data, close to 80% still use spreadsheets for data ingestion, cleansing, preparation and analysis. For a multitude of reasons, spreadsheets are not really the best tools in town for large time series or sensor data.
The reality of people working with data today is that more than 75% of their activities still pertain to data import, cleansing and preparation. Of course, data cleansing and preparation are quintessential for extracting value. But these activities are time consuming, mundane and do not generate the actionable insights that translate to value for the organization. These activities can include importing and synchronizing data from different databases with different data, time and date formats. It can mean accessing information from databases with different languages. If the dataset has different sampling times, then in most cases it is imperative to re-sample the data. These mundane tasks must be done in order to start any meaningful visualization and analysis of the time-series data from the sensors.
Once data has been cleansed and prepared, it is analyzed by experts who are well-versed with respect to the assets and the processes in the plant. They try to visualize anomalies in their data. They try to predict failures. They perform root cause analysis if a machine or a process exhibits anomalies. They try to get a feeling for the data based on the time waveform and variations. They look for parts of the data that are relevant and contains value and insights. For these analyses, they use tools available like Python, spreadsheets and MATLAB. The challenge is that they constantly straddle between machines, assets, codes and scripts. Since they are not constantly working with data science tools and programming, they tend to lose touch making it difficult to get started each time. Many of these experts are not programmers per se. Finally, there are not enough data scientists in the industry. It is expected that in the next decade, more than 50% of the data-related positions will remain unfilled.
The number of subject matter experts with data engineering skills is disproportionate to the number of critical assets that need to be analyzed and monitored regularly. These experts possess deep knowledge about the assets and the processes. They are amongst the most experienced and skilled people working at the plant. However, it is anticipated that in the next decade, more than 50% of this tacit industrial domain know-how would be lost to retirement. The knowledge base they’ve built over decades will leave the organization when they retire. This is because much of the knowledge about products, processes, assets and customers is not written down—it is in the workers’ heads. This knowledge unarguably is one of the driving force behind innovation and is critical to the company’s competitive advantage. The reality is that there are no tools at the disposal of the organization for knowledge sharing and collaborative decision-making on assets and processes.
For sure, digitalization is risky and requires courage. But not undertaking digitalization with the right tools is even riskier. Also, given the formidable challenges facing the industry, the consequences of not working with the right tools for digitalization can be staggering. The organization will miss out on their chances to improve production efficiency and machine uptime performance. The organization will miss the opportunity to transfer the captured expert knowledge to the next generation of engineers. Without digitalization, the organization may lose their competitive advantage. They become unable to collect key statistics about their assets, processes, products and customers. Businesses that don’t evolve and expand will find it difficult to retain market share. They will experience lower profitability given the increasing costs of industrial maintenance and production. Finally, early adopters of digitalization will have a learning advantage on the competition. Through testing and learning, organizations can get better insights, and update products and policies faster.
Viking Analytics provides self-service analytics software, MultiViz for the heavy industrial companies. We put the power of the data into the hands of the people who understand what it means: the subject-matter experts. MultiViz enables experts like condition-monitoring experts, reliability experts and process experts to rapidly visualize sensor data from assets and processes. It enables the experts to prepare, analyze and process sensor data without being a data scientist. Specifically, MultiViz enables the following.
- Experts can now explore large sensor data quickly and troubleshoot issues in assets and processes
- Experts can now monitor and contextualize process and asset data
- Experts can detect anomalies and irregularities in processes’ and assets’ performance
- Experts can identify operational/failure modes and detect abnormal mode changes
- Experts can rapidly annotate data for building ML models
- Experts can rapidly operationalize their insights as real-time applications
MultiViz helps the experts to contribute directly to overall plant and organizational profitability. It‘s human-centric design helps the organization to retain expert knowledge pertaining to critical issues and other irregularities. It enables seamless knowledge sharing and collaborative decision-making. In summary, MultiViz has been designed to help organizations to accelerate their digitalization efforts and alleviate risks in their implementation. Organizations can rely on data-driven methodologies for actionable insights and to deliver organization-wide value.