Toshiba presents new anomaly detection AI for large-scale industrial installations at ICDM2021 LITSA

Workforce transformation technology manages complex operating conditions through massive sensor data, contributes to efficient plant operation and maintenance.

TOKYO, December 7, 2021 / PRNewswire / – Toshiba Corporation (TOKYO: 6502) has developed a new anomaly detection AI that enables large-scale industrial facilities to meet a widespread and growing challenge: using a small workforce to ensure constant and efficient monitoring of thousands of sensors and the precise detection of abnormal signals hidden among small variations in sensor values. AI can currently be applied to power plants and industrial plants that use pumps to move fluids, and is the first of its kind* 1 to achieve high-level precision in the detection of anomalies in the complex interactions between the operating conditions of the plant and the massive values ​​of the sensors. Toshiba will present the technology at the 21st IEEE ICDM2021 International Conference LITSA Workshop on Data Mining on December 7.

Fig. 1 Technical overview of Toshiba’s new Anomaly Detection AI.

Fig.  2 Mikawa Factory in Omuta City, Fukuoka Prefecture, Japan.

Fig. 2 Mikawa Factory in Omuta City, Fukuoka Prefecture, Japan.

Fig.  3 The AI ​​for detecting anomalies in operation (demo).

Fig. 3 The AI ​​for detecting anomalies in operation (demo).

At the heart of AI is Toshiba’s “two-stage automatic encoder,” a proprietary deep-learning model that provides highly accurate predictions of sensor values ​​under normal operating conditions; it detects anomalies hidden in massive sensor data by identifying deviations between actual and predicted values.

Industrial facilities that have to move fluids with pumps, such as power plants and water treatment plants, use sensors to detect the values ​​of small and large fluctuations in operation. Relatively small and rapid fluctuations occur simultaneously on a few sensors due to pump vibration or local temperature change. The large fluctuations that occur on many sensors, with greater amplitude and slower cycle, reflect changes in power and plant operation.

Commenting on the new AI, Susumu Naito, Principal Investigator at Toshiba’s Corporate Research & Development Center, said: “The key factor in the success of this technology is the deep and extensive know-how that Toshiba has acquired through many years of experience in the automotive industry. energy and infrastructure. design of two deep learning models, one for each fluctuation characteristic, and guarantee of a very high level of precision in the prediction of the normal values ​​of the sensors. These are compared to the actual values ​​to detect anomalies. “

AI tests on open datasets from the Water Distribution (WADI) benchmark* 2 confirmed the highest level of detection accuracy in the industry, a 12% improvement over prior art* 3. In another test, Toshiba also verified that AI can recognize and report abnormal signs 6.8 days earlier than possible with manual monitoring by a trained operator. Early detection of anomalies enables condition-based maintenance and aids in the efficient operation and maintenance of the plant.

Toshiba conducts a demonstration experiment of online AI monitoring and early anomaly detection at the Mikawa power plant, operated by SIGMA POWER Ariake Corporation, a subsidiary of Toshiba Energy Systems & Solutions Corporation, in Omuta, Fukuoka, Japan (Figures 2 and 3).

In the future, Toshiba will prepare proof of concept (PoC) systems and explore applications on other types of industrial facilities. Once released, Toshiba plans to deliver AI both as an on-premise solution and as a cloud solution in the Toshiba SPINEX Marketplace, Toshiba’s IIoT service portal.


* 1 Toshiba poll.

* 2 Water Distribution (WADI): The scaled-down data set, including anomalies, of the actual water treatment plant.

PA Mathur and NO Tippenhauer, “SWaT: a water treatment testbed for research and training on ICS security,” Proceedings of the 2016 international workshop on cyber-physical systems for smart water networks, pp. 31-36, april 2016.

* 3 Machine learning techniques for anomaly detection such as Unsupervised Anomaly Detection (2020) and OminAnomaly (2019). The comparison of each of these methods is discussed in the following document.

S. Naito, Y. Taguchi, K. Nakata, Y. Kato, “Anomaly Detection for Multivariate Time Series at a Large-Scale Fluid Processing Plant Using a Two-Stage Automatic Encoder.”

About Toshiba Corporation

Toshiba leads a global group of companies that combine the knowledge and capabilities of over 140 years of experience in a wide range of businesses, from energy and social infrastructure to electronics, with world-class capabilities in health technologies. information processing, digital and AI. These distinctive strengths support Toshiba’s continued evolution to become an infrastructure services company that promotes the use and digitization of data, and one of the world’s leading cyber-physical systems technology companies. Guided by the Toshiba Group’s core commitment, “Committed to People, Committed to the Future”, Toshiba contributes to the positive development of society with services and solutions that lead to a better world. The Group and its 120,000 employees around the world have generated annual sales exceeding 3.1 trillion yen (US $ 27.5 billion) in fiscal 2020. To learn more about Toshiba, visit

Toshiba logo (PRNewsfoto / Toshiba)

Toshiba logo (PRNewsfoto / Toshiba)

SOURCE Toshiba Corporation

About Emilie Brandow

Check Also

Leasing Market Size and Forecast to 2029

Los Angeles, USA,-According to the report of the verified market report, the global leasing market …