Industrial Big Data and Its Value

As the curtain slowly rises on a new era of industrial revolution, technologies such as the Internet of Things (IoT), the Industrial Internet, intelligent ICT, and artificial intelligence are becoming the brightest new stars on stage. Driven by these emerging technologies, an industrial big data environment is gradually taking shape: data is shifting from being a by-product of manufacturing processes to a strategic resource that enterprises highly value.

If big data is the core driving force behind the transformation of industrial value, then how should we define and use it? In our book Industrial Big Data, we expressed a view on this question: big data is not an end in itself, but rather a perspective for analyzing problems and a means for solving them. By gaining insight from data, we can forecast demand, predict manufacturing, mine the value of the invisible world, address and prevent hidden risks, and integrate the industrial and value chains through data. These constitute the true core value and purpose of big data.

What Is Industrial Big Data?

When big data is mentioned, people first tend to think of applications in Internet and commercial contexts, where large amounts of behavioral data are used to analyze user behavior and predict market trends. But the definition and application of industrial big data are much harder to intuitively understand and imagine.

The most popular definition of big data today comes from the “4V” characteristics proposed in Big Data: A Revolution That Will Transform How We Live, Work, and Think by Viktor Mayer-Schönberger and Kenneth Cukier: Volume, Velocity, Veracity, and Variety. This definition targets big data in Internet and social environments and is framed in terms of the technical challenges of data engineering.

The challenges and objectives of industrial big data, however, are better understood through “3B” and “3C”.

The “3B” challenges of industrial big data applications:

Bad Quality — In industrial big data, data quality has long been a major challenge for many enterprises. This is mainly constrained by the limitations of data acquisition methods in industrial environments, including sensors, data acquisition hardware modules, communication protocols, configuration software, and other technical constraints. Mastery of data quality management is a hard skill every enterprise must develop.

Broken — Industry’s requirement for data is not only about volume, but more importantly about completeness. When using data modeling to solve a specific problem, it is necessary to obtain a complete set of parameters related to the object of analysis. The absence of certain key parameters can fragment the analysis process. For example, analyzing the performance of an aircraft engine requires multiple parameters such as temperature, air density, inlet and outlet pressures, and power. If any one of these parameters is missing, it becomes impossible to establish a complete performance evaluation and prediction model.

Therefore, before data collection, enterprises must clearly plan their targets and analytical objectives to ensure the completeness of the data they obtain, so as to avoid discovering, after investing heavily in data accumulation, that the data cannot address the problems they care about.

Below the Surface — In addition to analyzing the surface-level statistical characteristics reflected by data, we must also focus on the correlations hidden within it. Analyzing and mining these beneath-the-surface correlations requires comparative, reference-type data, a process known in data science as “labeling.” Such data includes operating conditions, maintenance records, task information, and so on. Although the volume of this data is usually small, it plays a crucial role in data analysis.

The “3C” objectives of industrial big data analysis:

Comparison — Gaining insight through comparison, including both similarity and difference. The dimensions of comparison may involve comparing the same object over time (temporal dimension) or comparing one object to other individuals in a cluster (population dimension). Such comparative analysis helps us classify large volumes of individual information and lays the foundation for identifying universal rules within similarities and causal relationships within differences.

Correlation — If the Internet of Things connects the visible world, then the correlations among connected objects constitute the connections of the invisible world. Mining correlation is the basis for forming memory and knowledge. Simply storing information does not constitute memory; memory fundamentally consists of managing information through associations and using those associations to spark heuristic recall.

Correlation also enhances the efficiency of the human brain in managing and retrieving information. When we recall a scene or an episode, we usually do not remember every detail. Instead, we pull on a “thread” of a clue, and this clue brings the entire scene back to mind. Applying this memory-like information management approach in industrial intelligence yields a more flexible and efficient method of data management.

Consequence (Causality) — A major goal of data analysis is to support decision-making. When making a specific decision, the results and impacts it brings should be equally analyzed and predicted. This is something that conventional control systems have not possessed, and it is also the essence of intelligence.

Most activities in industrial systems are highly goal-oriented, aiming to maximize target performance and minimize damage, i.e., “result management.” Prediction is the foundation of result management. For example, in a manufacturing system, if we can predict the impact of equipment degradation on quality, as well as its impact on the quality of subsequent processes, we can compensate for and manage quality risks during production. As a result, the manufacturing system becomes more resilient and robust.

In summary, Internet/commercial big data and industrial big data differ significantly in terms of technical challenges, data attributes, and analytical objectives, and these differences in turn determine the divergence in technical approaches.

The Value and Significance of Industrial Big Data

In the process of industry and commerce evolving from separation to integration, we have observed four important transformations worth attention: changes in customer needs, changes in production systems, changes in business models, and changes in industrial system operation modes (decision-making modes).

Changes in customer needs have progressed from “none to some,” then “some to refined,” and now from “refined to personalized.” Changes in production systems have evolved from simple to complex, from complex to large-scale, from large-scale to lean, from lean to flexible, and from flexible to intelligent, with the ultimate goal of achieving a worry-free production environment.

Changes in business models have moved from selling products to selling capabilities, from selling capabilities to selling services, and from selling services to selling value, with the goal of uncovering users’ invisible value.

Changes in industrial system operation modes have moved from reactive problem-solving, to experience-based problem prevention, and are now shifting—driven by data—to problem avoidance based on clues and facts, ultimately aiming to realize the acquisition and inheritance of knowledge.

At the core of this whole series of transformations lies the transformation of industrial value, and big data will play a pivotal role in this process. Put simply, the goal of industrial big data is to enable the progression from “making by oneself” to “manufacturing,” from “manufacturing” to “intelligent manufacturing,” and from “intelligent manufacturing” to “knowledge inheritance.”

Using Big Data to Mine Customer Value in the “Invisible World”

“One uses what is there for profit, and uses what is not there for utility” is a line from Laozi’s Tao Te Ching, and its embedded wisdom still applies perfectly to today’s industrial value models. It can be understood as follows: the tangible entity of any object provides us with visible foundational conditions to build upon, but the hidden space within it and its boundless potential for change are where we truly use and create value.

In Industrial Big Data, we used a “fried egg model” to describe the relationship between product value and service value (see Figure 1): the yolk represents the product itself, whose differentiation and degree of customization are often not pronounced. For instance, if you cover the logo on a TV set, it becomes difficult to distinguish which company produced it.

The egg white, by contrast, represents value-added services, which are a crucial embodiment of differentiation and customization, and constitute the core of a company’s brand and sustainable value. These values reside in the “invisible world” of user usage scenarios, the correlations among hidden factors, and the full life cycle of product manufacturing and use.

Data will become a key means of mining these values, mainly in three ways:

• Using data to extract new knowledge and technologies from usage and apply them to improve existing products;
• Using data to discover and define users’ latent, as-yet-unknown needs;
• Using data as a medium to provide value-added services to users.

Take wind power generation as an example. The differentiation among wind turbines themselves is not very prominent, nor are users’ customization requirements particularly strong. However, power generation capability, operational stability, and operation and maintenance (O&M) costs during turbine operation are at the core of user value.

By leveraging operational big data, we can carry out health management for wind turbines, predict potential operational risks, and optimize wind farm O&M. This in turn improves turbine availability, enhances power generation efficiency, and reduces O&M costs. Turbine manufacturers no longer have to rely solely on one-off profits from selling equipment; they can also generate ongoing profits by providing value-added services during product usage.

Data-driven intelligent O&M services have thus become a key strategic direction for many enterprises seeking value transformation.

The IMS (Intelligent Maintenance Systems) Center, in collaboration with Nissan, introduced industrial big data predictive analytics models for health management of industrial robots. Because industrial robots are widely used in highly complex production environments, adding external sensors is often not appropriate. Instead, health analysis is performed using monitoring parameters from the robot controller.

A significant portion of Nissan’s industrial robots are six-axis robotic arms; failure in any axis will lead to downtime of the entire arm. After using servo-axis rotational speed signals to distinguish among different operating modes of the robotic arm, health assessment models are then built for condition parameters within each mode (such as torque and temperature). Analysis revealed that early fault characteristics could be predicted as early as three weeks before failure occurred (see Figure 2).

Nissan subsequently began promoting predictive analytics models across its six-axis servo robotic arms. After collecting and analyzing large amounts of robot data, they conducted clustering analysis on robotic arms of different types and under different operating conditions, forming “virtual communities” of robots. Using cluster modeling methods, they analyzed the data of each robot community; by comparing each arm with its cluster peers, they could determine its level of abnormality and rank the health status of all robots within the cluster.

After quantifying the health status of the robotic arms, Nissan networked the analysis results through content management and established an online monitoring system for a “virtual factory.” In this virtual factory, managers can perform multi-level, vertical management of equipment status—from the production system level, to the production line level, workstation level, individual machine level, and even down to key component level. Maintenance and production plans can be scheduled according to the real-time condition of equipment.

The system can also generate a daily health report, ranking and statistically analyzing the health status of all equipment on the production line. It provides equipment managers with the health risk status and main risk locations for each device. This enables more focused daily inspections—no risk point is overlooked, yet unnecessary checks and maintenance work are minimized—achieving a transition from preventive maintenance to predictive maintenance.

After more than 200 years of technological revolution, human society has amassed a huge stock of industrial products. The basic infrastructure of industry and many fundamental production factors—such as machine tools, power facilities, power-generation equipment, manufacturing equipment, and transport equipment—are approaching saturation in terms of demand.

Take Germany as an example: its industrial export output value showed no growth for six consecutive years starting in 2006. The root cause is that developing countries have gradually completed industrial upgrading, and their demand for industrial equipment is largely saturated. Consequently, Germany’s “Industry 4.0” strategy emphasizes integration and software services for manufacturing systems, manifested in “vertical integration,” “horizontal integration,” and “end-to-end integration.”

GE in the United States has observed the same phenomenon. The company realized that profit from equipment sales is far less than the value that can be created by services during product usage. Customers need more than maintaining product condition; they care about how to use these capabilities to achieve more efficient value creation.

Putting data at the core to unlock a product’s maximum capability ultimately comes down to using data modeling to accurately evaluate state, environment, and tasks; to conduct real-time decision optimization for management and control activities; and to coordinate and schedule related products so that they operate at maximum efficiency.