Jump to the main text

Hitachi

Hitachi America, Ltd.Hitachi America, Ltd.

Hitachi America, Ltd., Research & Development

Big Data and Analytics Solutions

Analytics Horizontals -- AI and Machine Learning for Industrial IoT

Big Data and Analytics Solution

Hitachi enjoys the leadership across a diverse portfolio of Operations Technologies (OT) and Information Technologies (IT) in multiple industry segments. This leadership position is further enhanced by the convergence of the IT and OT portfolios through the leverage of Big Data and Analytics Solutions that addresses the entire spectrum of data collection and control systems to enable end to end solutions. Hitachi’s solutions are innovation led and address hard business problems recognized by the participating industries. The following portfolios illustrate some of the early successes and applications of the Big Data and Analytics Solutions.

Oil & Gas

Use Case (1) - Formation Top Identification

Formation tops are the point at which the age of one rock sequence starts and another ends. Accurate identification is important for a number of economic, commercial and safety reasons. For example, during well planning and design, data from previously drilled wells (offset data) and seismic exploration models (limited accuracy) identify the point at which casing is set. Casing the well involves running wide bore tubular steel ‘casing’ into the previously drilled hole then cementing this in place. Wells use casing to stabilize the well, or when entering a formation where the surrounding pressures are significantly different and new lubricating mud is needed. Casing points are aligned (usually) with formation tops marking changes to the geology, so accurate identification as close to real time as possible is desirable. Selecting formation tops is, however, subjective with the real time data showing variations in pattern from one well to another, and as such is usually confirmed once cuttings (rock chips from the drilling process) are recovered and inspected by a geologist.

The main challenges resulting from this manual formation top identification (FTI) are:

  • Non-productive time incurred due to inappropriate casing decisions
  • Potential Health, Safety & the Environment hazards from unexpected kicks/pressure variations
  • On-time finishing of the drilling operation

The main motivation of our partner(s) comes from the fact that many of their clients have been actively drilling a large number of wells but often missing critical formation tops. The ideal expectation is to solve this problem through achieving two main objectives:

  • Automate the identification and notification of encountering a formation top of interest as soon as it is reached.
  • Remove as much subjectivity in identification to provide consistency across all wells in a particular basin.

Formation top Identification application

We achieved significant accuracy in FTI results by applying machine learning algorithms on the sensor measurements from LWD/MWD systems in drilling.

Use Case (2) - Remote Services Management for Natural Gas Compression Service:

Global energy consumption increased from around 11,296 MTOE (Million Tons of Oil Equivalent) in 2010 to 12,034 MTOE in 2014. Moreover, the share of natural gas in the total global energy mix increased from 25.3% in 2010 to 25.5% in 2014. Hence, backed by growing consumption of natural gas, the demand for natural gas compressors is anticipated to grow at a robust pace over the next five years. With over 2M wells in NA, the adoption of sensor instrumentation and monitoring services of these assets is steadily growing.

There are a large number of small to medium size service providers addressing these markets for uptime of these CAPEX heavy assets. These service providers are challenged by operational inefficiencies due to the current age disparate systems which lack the context of the process and ecosystem the assets operate in to improve operational efficiencies. Using Hitachi’s innovative big data analytics combined with the services of our partner(s), we plan to disrupt the industry practices in improving overall operational efficiencies by up to 30%. This type of holistic sensing and big data analytics integrated into the solution allows Hitachi to expand its businesses in the IOT space through the long tail integration of services.

Key Analytics Functions includes:

  • Detect compressor shutdowns and identify false alarms
  • Identify operating envelopes and process characteristics based on reservoir conditions.
  • Manage operational efficiencies by improving Non Productive Time (NPT)
  • Compressor failure prediction analysis
  • Asset optimization: Prediction models to anticipate maintenance schedule, material and labor force scheduling to optimize productivity of the assets and equipment.

Remote Services management for Natural gas compression

Agriculture

Farm-to-Fork Value Chain Integration

We focus on enhancing the entire value chain from farm to fork using robust measurement framework, technology interventions, and big data analytics. We optimize the crop portfolio by enabling crop models, farm information model, soil model, integrated nutrition and pest management. For the measurement framework, we have prototyped and deployed an IOT network based on ultra-low power sensor nodes calibrated for real farm conditions. These sensors measure different soil characteristics at various depths, including temperature, moisture, PH, and electrical conductivity among others. We have set up an Agro- Advisory Service, a cloud-based analytics solution which enables crop portfolio monitoring. This is also integrated into a mobile application to aid on-field measurement and diagnostic Services.

Framework for Agro-advisory solution

Sensing and Agronomy Data Collection Scheme

Sample screenshots from Web and Mobile Apps

Energy

In the past decade, the Power and Energy Industry experienced great advancement of communication and electronics technologies, which accumulated massive amount of structured and unstructured data with unforeseeable complexity and velocity. While thrilled by increasing information availability, utilities, vendors and end users are also overwhelmed by a variety of data silos, incompatible data protocols and lack of advanced analytics. Mostly, they struggle to convert data to useful information and find it challenging to bring traditional business intelligence to the next level.

The Energy Solutions team targets to solve customer challenges on data collection, management and analytics. We stand out with our customer-focused innovation approach, particularly in the following areas:

  • Sensor data collection and ingestion
  • Cyber-secured data management on Hitachi cloud or proprietary servers
  • Advanced data analytics on
    • Generation and load forecast
    • Optimal control and diagnosis of energy storage systems
    • Abnormality detection for time series sensor data
    • Data driven distribution asset management and optimization

PMU data facilitated grid oscillation

Advanced Asset Management App

Power Grid Reliability Outlook App

In the area of renewable energy research, the energy solution team is leading a customer-driven co-creation research project with a major renewable power producer in India.

We are currently focused on providing solution for the day-ahead power generation forecasting for wind turbines, with an eye toward photo-voltaic systems and hybrid systems in the near future. Values of creating a forecasting solution-core for renewable energy solutions are:

  1. Near term: independent power producers need to have accurate visibility to 24-36 hour look ahead forecasting for selling to utilities
  2. Long term: (a) energy trade pricing optimization and (b) short term generation and demand flexing as in the case of Micro-grids.

Our forecasting technology was recently put to test at the open forecast competition organized by in the International EEM2017 Conference in Germany and was awarded third place winner.

We also partner with the Oak Ridge National Laboratory on Sunshot, a research project sponsored by the US Department of Energy aimed at improving solar forecasting. More specifically, the research concentrates on near-range solar irradiance reforecasting based on regime identification and weather state forecasting with improved WRF-Solar model and data assimilation.

We recently engaged with a major energy distribution company to develop analytics for load prediction of a large network of power distribution transformers using smart meter data.

Finally, in the area of Microgrid research, we collaborate with Omika Works factory in Japan to research how to use big data analytics to manage energy cost through optimal control of resources deployment such as optimal battery charge/discharges. More specifically we develop solutions for energy IOT, demand forecasting, solar forecasting, and dynamic control optimization of battery operations.

Fleet Management

In automotive industry, as the penetration of telematics system and Advanced Driver Assistant System (ADAS) is increasing, many cars and trucks are getting connected with different kind of infrastructures and other vehicles. This is creating an opportunity to create new services in the mobility space for passenger car and commercial fleets alike. These services include advanced fleet management system (FMS), autonomous driving solutions, and usage based insurance (UBI). By bringing together connectivity, the latest in AI and machine learning techniques and design based thinking, the service providers can optimize the transportation system, provide more convenient services, and reduce costs for driver wages, fuel, and maintenance drastically.

At BDL, we are developing a distributed digital fleets solutions platform to create new ITxOT solutions for fleet companies. Some of the key technologies of this platform are:

  • Edge device for data collection and in-vehicle analytics;
  • Connectivity platform capable of efficient transfer of data from the edge devices to the core platform;
  • Core analytics platform for advanced data analytics;
  • Advanced analytics enhanced with high fidelity simulation of component level failure prediction using physical-based simulation;
  • Driver and vehicle monitoring, profiling, and performance analytics.

Healthcare

Large amounts of data are being generated from all segments of healthcare. From image data coming from MRI machines and CAT scanners, to clinical data from Electronic Health Records, to device data from ICU devices, to financial data from payers, all the way to personal health data from wearables. The Big Data Laboratory (BDL) is looking at harnessing all this data, structured and unstructured, real-time and historical, to develop advanced analytics technologies to help Hospital Operations Management, Accountable Care Organizations (ACOs), Population Health Management, Comparative Effectiveness Research, Precision Medicine, Patient Monitoring, Clinical Decision Support Systems, and many other forward looking areas.

To help in these different areas of healthcare, the Big Data Laboratory (BDL) is creating key technologies to construct deep clinical signatures from patients, to match patients to past patients that had similar signatures, and to make key predictions based on what happened to those past patients. These predictions are then used to streamline the operations and the clinical decision making across the healthcare system.

Healthcare applications

To help hospitals streamline their operations and become more efficient, we are creating a cloud based solution to make key predictions at key decision points. The various applications that make up this solution are described below.

Application Description
Admission and Discharge Analytics
index_clip_image008_0000.png

From the time patients are admitted to when they are discharged, we assess various metrics to help make key clinical and business decisions:

  • Length and cost of stay
  • Risk of hospital-acquired conditions
  • Risk of falls
  • Mortality
  • Risk of readmission within 30 days
  • Risk of ER visit within 30 days and within 6 months

These key metrics are updated with new test results, new diagnoses, and other new information during the patient’s stay.

Example applications:

  • A patient with a high risk of a hospital-acquired condition is isolated from the start, and her doctor is informed
  • Similarly, for a patient with a high risk of falling, the nursing staff is alerted
  • A patient with a high risk of 30-day readmission can be discharged to the appropriate facility or to homecare, with the appropriate care and monitoring
Care Path Analytics
index_clip_image010_0000.png

During a patient’s hospital stay, as more clinical information, such as lab test results, becomes available, we predict the possible sequences of medical interventions based on similar patients. This can help doctors and hospitalists make key clinical decisions and streamline hospital operations.

Example applications:

  • The doctor can choose the next procedure or path along a pathway that is associated with a good trade-off between readmission risk, other risks, and cost
  • The hospitalist can aggregate the predicted resource demand from individual patients to better match supply with demand, for medical resources like nursing staff, beds, and medical equipment
  • Serves as a hindsight analytics tool to discover frequent care path patterns within a group of similar patients (such as a Diagnosis-Related Group) to reduce unnecessary variations, improve care quality, and reduce costs
ICU Analytics
index_clip_image012.png

We display an ICU patient’s device data and nurse observations and use this data to predict key short- and long-term risks.

Example applications:

  • We monitor a patient’s predicted risk of needing a mechanical ventilator in the next several hours and alert the staff as needed
  • For a patient on a ventilator, we monitor her readiness to be weaned and notify the staff as needed. This reduces unnecessary time on the ventilator, and hence reduces the risk of ventilator-associated pneumonia.
Radiology Analytics
index_clip_image014.png

When patients take imaging examinations, we provide key radiology analytics to help radiologists and hospital staff make clinical decisions:

  • Lung nodule detection for low-dose CT images
  • Vessel suppression for low-dose CT images
  • Text analytics for radiology reports
  • Coil performance monitoring for MRI coils

Example applications:

  • A patient reporting chest pain took a screening CT scan, and the radiologist is provided the detected lung nodules (with background vessels suppressed) in the screening scan
  • A clinician is presented with keywords extracted from the patient’s historical radiology reports
  • Hospital maintenance staff is informed about MRI coil performance from daily calibration MRI scans
Wearables Analytics
index_clip_image016.png

"Today’s ICU device is tomorrow's wearable"

The wearables available on the consumer market are becoming more sophisticated and accurate. For less than $100, one can buy an activity tracker that measures heart rate variability (HRV), which is associated with one’s stress level.

Examples application:

  • Personal health monitoring for casual users and for those on home care
Predictive Cohort Analysis – Our underlying technology
index_clip_image018.png

Predictive Cohort Analysis is the underlying platform for end-to-end individual and population health management.

As an individual health management tool: Once a patient is admitted, we construct a signature using deep and wide data, and provide a holistic clinical view of the patient’s risks and best care plans using her personalized cohort of similar patients. By tracking patients from admission to post-care, it offers better, more efficient care to patients while reducing costs.

Example applications:

  • Patient similarity analysis can be used to provide clinical decision support. Doctors can tailor individual treatments or order appropriate lab tests for patients. Also, they can flag patients with care gaps, who require more attention due to poor response to therapies, missing procedures, or lab tests.
  • Predict a patient’s prognosis or trajectory over time by leveraging inter-patient similarities

As a population health management tool: We provide various ways to visualize, create, expand, and alter cohorts. The resulting cohort can be used for clinical trials and comparative effectiveness studies.

Example applications:

  • Domain experts can perform interactive, exploratory cohort studies to uncover correlations between specific risk metrics and the underlying attributes of individuals within the study population
  • Doctors can build powerful analytical cohorts of people that are likely to use the emergency room in the next 30 days, and reach out to them with preemptive intervention

Analytics Horizontals - AI and Machine Learning for Industrial IoT

We are at the cusp of transformative changes across industries, from agriculture to manufacturing, from mining to energy production, from healthcare to transportation. These transformations hold the promise of making our economic production more efficient, cost effective and sustainable and are being driven by the convergence of the global industrial system (Operations Technology) with the power of integrating advances in AI and machine learning, advanced computing, low-cost sensing and new levels of connectivity (Information Technology). This convergence enables the creation of a new class of big data solutions for monitoring, managing, and optimizing industrial operations and physical systems.

At Hitachi, we realize that there is a need for such solutions in the broader marketplace. We are using our decades of experience in equipment manufacturing and operations and marrying it with our expertise in analytics to bring unique solutions to market that solve some hard and important customer problems. Our extensive experience in working closely with customers in addressing their challenges has led us to understand and develop a unique set of horizontal technologies that are solving hard customer problems in the industrial world. Some of the horizontal areas we are working on are:

  • Predictive Maintenance
  • Improving Quality
  • Operations Optimization
  • Increasing Safety
  • End-to-end Control and Optimization
  • Analytics Framework and Environments for Industrial Applications

Predictive Maintenance (PdM)

The Big Data Laboratory (BDL) is developing a suite of technologies to address predictive maintenance use cases for solutions across verticals. Current focus is on understanding customer requirements, and developing solutions that address these requirements. The BDL data-driven technologies enable the maintenance staff to take the right maintenance actions at the right time, resulting in the avoidance of unexpected failures, decrease in maintenance costs, and increase in equipment availability.

The developed PdM technologies employ AI and machine learning techniques over historical event and sensor data in order to:

  • monitor equipment,
  • detect performance degradation,
  • estimate maintenance effectiveness,
  • analyze root cause of failure or degradation,
  • predict future failures,
  • predict remaining useful life of equipment,
  • recommend operating envelopes,
  • recommend maintenance and repair actions, and
  • optimize the maintenance process.

The developed technologies have use cases in a variety of verticals, including but not limited to: manufacturing, transportation and automotive, infrastructure management, oil and gas, mining, and health industry.

Operations Optimization

Based on our extensive research experiences in many operational areas such as fleet management, mining, rail, manufacturing, etc., we are building suite of horizontal technologies to optimize operations and enhance productivity. We are using a combination of AI, machine learning, simulation and optimization to enhance utilization, overall equipment effectiveness, while reducing the cost of operations.

Some of the key technologies we are working on are:

  • Production monitoring,
  • Analysis of operators’ behaviors,
  • KPI forecasting,
  • Operation parameter recommendations
  • Equipment scheduling
  • Dynamic dispatch

Supply Chain Analytics

Big Data and Supply-Demand Analytics

Research Solutions and Development of a demand-supply matching tool for trading companies by enabling the following:

  • Data load and user based research to correlate external data to demand from any vertical.
  • Planning, procurement, and forecasting. Cloud services for any external client. It allows for clients to use it with their own data while the sales/negotiation of Analytics deal process goes though.
  • This work is part of new services in Business Development.
  • Development of Solutions of demand forecast of parts, tools, and consumable products
  • Development of probability predictions of winning opportunity deals of expensive, low volume sales equipment. Addressing problems of measuring win probability in sales efforts for deal management.

Supply chain Analytics and Optimization for Tubular products:

  • Develop demand/supply analytics under block-chain architecture to manage and track transactions of tubular assets from Mills to oil wells. We are developing prediction models to explore.
  • Research work in this space includes predicting failure events of the wells through survival models, and detecting from various signals best timing to preempt service disruptions and replacements of pipes.
  • Model and forecast tubular product demands from oil and gas projects, uncertain due to the nature of oil exploration, location, potential depth of deposits, etc.
  • Because of high costs of pipes shortages, it is common practice to over-supply them, causing waste and potential loss of assets. In this context, we develop analytics of buy-back contracts models, returns policies, dynamic procurement policies, as well as trans-shipment policies to optimize the asset utilization.

Asset optimization: In CAPEX-heavy industries, expensive equipment show multiple failure modes which can disrupt and create costly down time on products, as well as disrupting demand-supply logistics. We build prediction models to anticipate maintenance schedule, material and labor force scheduling to optimize productivity of the assets and equipment.

Use Case - Supply Chain Optimization for OCTG

Asset lifecycle management is extremely critical for industries with CAPEX-intensive assets. Specifically, in oil and gas sector, there are numerous applications. One example involves managing the rig, where a significant amount of capital is spent on rental/contracts. Another example involves the tubulars which account, on an average, for over 10% of the cost of typical oil well. The tubular supply chain business has a large number of entities with disparate enterprise systems which are very costly to integrate in a point-to-point manner. There is an opportunity to disrupt this market by providing a single, consensus-based integral solution to manage the movement of tubulars from manufacturing source to well site.

The proposed single truth system will enable the following:

  • Smooth integration of the enterprises across the supply chain.
  • Enabling edge devices to measure and monitor IOT data.
  • Leveraging IOT data with Blockchain technology (https://www.hyperledger.org/)
  • Integration of IOT data with Rig-In-a-Box micro application services.
  • Analyze demand and supply signals tuned with market.


Blockchain analytics enabled through Micro-application services

Common Analytics Framework

There is a strong customer demand for complex big data analytics applications across vertical such as energy, healthcare, automotive, and mining. One approach to building such applications is to build them from scratch for every customer. However this approach has the drawbacks as follows.

  • Inefficient, resource intensive, expensive
  • Hard for domain experts to be involved in the development, hence incomplete
  • Loss of opportunities to leverage learnings from other applications

This means that we need a different approach, which is shown in Fig.1 on the right hand side (the left hand side shows the first mentioned approach). By building a common analytics framework, we will be able to significantly reduce application development and deployment time.


Two approaches to building analytics applications for a number of verticals

The framework consist of front end that empowers application builders easily to construct analytics logics as well as dashboards and back end that can take care of multiple execution platforms for analytics operations extending from simple to complex, such as machine learning algorithms for predictive maintenance. The framework enables domain experts to build applications and it can keep incorporating “knowledge” from a new applications.

Video Analytics

Hitachi has performed video analytics research for more than a decade, and we have a variety of technologies that are applied on compelling real-world use cases. An example of such an application is our automotive stereo camera system, which performs pre-collision breaking and throttle management to improve safety.

Our unique technology for similar image search can search through ten million images per second on a single server. Our Gazopa web site, which we used to demonstrate it, was selected as a TechCrunch50 finalist in September 2008.

By combining various object recognition technologies and our search platform we can create solutions for various verticals such as public safety and media and entertainment. For example, person detection and recognition using the latest AI technologies enables high-speed person search and tracking for public safety applications. (http://www.hitachi.com/New/cnews/month/2017/03/170327.html).

List of Publications

  1. Dayal, U., Akatsu, M., Gupta, C., Vennelakanti R., Lenardi, M. (2014). Expanding Global Big Data Solutions with Innovative Analytics. Hitachi Review, 96(4), 259-265.
  2. Gupta, C., et. al. Analytics-Driven Industrial Big Data Applications. Journal of Japan Industrial Management Association, 25(2), 96-105.
  3. Dayal, U., Gupta, C., Vennelakanti, R., Vieira, M., & Wang, S. (2015). An Approach to Benchmarking Industrial Big Data Applications. In Proc. of Workshop on Big Data Benchmarks (WBDB 2014), 45-60.
  4. Gupta, C., Farahat, A., Hiruta, T., Ristovski, K., & Dayal, U. (2016). Collaborative Creation with Customers for Predictive Maintenance Solutions on Hitachi IoT Platform. Hitachi Review, 65(9), 403.
  5. Zheng, S., Ristovski, K., Farahat, A., Gupta, C. (2017). Long Short-Term Memory Network for Remaining Useful Life Estimation. In Prognostics and Health Management (PHM), 2017 IEEE Conference on. IEEE.
  6. Ristovski, K., Gupta, C., Harada K., & Tang, H-K. (2017). Dispatch with Confidence: Integration of Machine Learning, Optimization and Simulation for Open Pit Mines. International conference on knowledge discovery and data mining (KDD). (To appear)