FEATURED ARTICLES Manufacturing Solutions Assisting Quality and Safety

Use of AI to Enhance Steel

Production Value Chain
Hitachi offers a wide variety of solutions to the challenges posed by rationalization,
operational efficiency improvement, and other issues facing customers in the steel and
non-ferrous metal industries. This article describes three of these solutions in terms
of how they use data and AI. Planning Optimization Service MLCP is used to improve
operational efficiency through the optimization of plant production plans, HITSODAS
is used to improve quality and yield through the analysis of large amounts of data
from high-speed control systems that operate on timescales in the order of tens of
milliseconds, and the Hitachi Motor Predictive Diagnosis Solution uses a patented Hitachi
algorithm for the analysis of motor current signals to improve the efficiency of electric
motor maintenance.

Tsutomu Ito
Zhe Cao
Kenji Maegawa
Naoya Eimaeda
Taku Saito

Diagnosis and Analysis System (HITSODAS) is used to

improve quality and yield through the analysis of large
1. Introduction amounts of data from high-speed control systems that
operate on timescales in the order of tens of milliseconds,
Hitachi has a track record of supplying electrical machinery and the Hitachi Motor Predictive Diagnosis Solution
control systems for the equipment used in the production of improves the efficiency of electric motor maintenance by
steel products such as high-quality automotive sheet, work means of predictive diagnosis that uses a patented Hitachi
that has involved it in the optimization of different types algorithm to analyze motor current signals (see Figure 1).
of equipment. In recent years, however, the steel industry
has been faced with falling prices for its products due to 2. Optimization of Production
oversupply and the resulting supply-demand imbalance, Planning Using MLCP
with the result that Hitachi’s customers in the steel and
non-ferrous metal industries are under pressure to rational- As production planning (scheduling) in the steel industry is
ize and improve operational efficiency. a task that only a small number of highly experienced staff
Hitachi offers a wide variety of solutions that address these are able to perform, its reliance on particular individuals
customer challenges through the plant-wide optimization has become a cause for concern. Moreover, the high level
of customer operations. This article describes three of of difficulty means that the work has resisted attempts
these solutions in terms of how they use data and artificial to computerize it by conventional methods. In response,
intelligence (AI). Machine Learning Constraint Programming Hitachi is seeking to overcome this challenge through
(MLCP) is used to improve operational efficiency through the use of its Hitachi AI Technology/MLCP, a planning
optimal plant scheduling, the Hitachi Self-organized optimization service.

100.
Figure 1 — Steel Mill Layout
The diagram shows the solutions available from Hitachi for each step in the production process, equipment is lined up from the unloading of raw materials
from ships on the left to product dispatch on the right. The thick lines indicate areas that are a particular focus for Hitachi.

Raw material yard Hitachi Motor Optimization of

Vessel Planning Optimization of
instrumentation, Energy
location Predictive Optimization logistics and traffic warehouse
optimization of efficiency
data Diagnosis Solution Service (MLCP) management management
inventory management

Raw Hitachi Self-organized

materials Diagnosis and
Analysis System
(imported) Dispatch
(HITSODAS)

Plan and
Hot actual data
Optimization of
Thermal power Measurements
on-site logistics
Blast rolling
generation Cold
furnace
rolling
Steel-making
Inspection
Drone techniques Transfer and use
of operational Carrier pallet
know-how
Materials yard Maintenance
support

Heavy Torpedo car Product yard

machinery

Centralized management
Hitachi
of production records, Support for
Ultra-high-speed
operations, worker safety
database engine
personnel, and so on

MLCP: machine learning constraint programming

The reliance of production planning on particular indi- through the use of MLCP to enable on-the-spot generation
viduals can be attributed to the following three reasons. of optimal plans (see Figure 2).
(1) High level of difficulty The main features of the system are as follows.
Planning needs to take several dozen considerations (1) Machine Learning
into account. Along with equipment constraints, these This involves learning how to replicate the know-how
also include materials, delivery times, and customer-specific of experienced staff by analyzing plan data that they have
requirements. There are also cases where not all conditions prepared. In particular, this provides a similar level of
can be satisfied, meaning that planning needs to settle for planning flexibility to that of experienced staff by replicating
a certain level of quality in which some conditions are their expertise in how to relax constraints when not all
deliberately loosened. constraints can be satisfied. The system also updates the
(2) Frequent plan revisions know-how of experienced staff by continuing to learn from
Issues such as changes to orders or problems with planning data after installation.
equipment arise on a daily basis, each time requiring plan (2) Mathematical Optimization Engine
revisions. Revising plans quickly without compromising The optimization engine delivers higher quality planning
their quality calls for the judgement of experienced staff. by identifying which plans offer the best performance in
(3) Inability to document know-how terms of the evaluation criteria while still satisfying multiple
Because steel is produced to order, the product variations constraints.
are complex with a wide variety of grades and sizes. This Along with this IT, there is also scope for further improving
makes expertise difficult to document. Updating staff know- plan accuracy by acquiring the necessary elements such as
how is also difficult given how the requirements change operational data, etc. from operational technology (OT)
with changes to equipment or to the product mix. control systems.
To address these issues, Hitachi has been using big data With MLCP, customers are able to produce high-quality
analytics to replicate the production plans of experienced plans quickly and overcome the problem of planning being
staff, working on the computerization of planning work so dependent on the expertise of particular individuals.

Hitachi Review Vol. 70, No. 2 212–213 101.

 Figure 2 — Block Diagram of
Planning Optimization Service
Hitachi AI Technology/MLCP MLCP
The service uses big data analytics to
replicate plans produced by experts. Understanding of
Optimal plans can be produced as business process
A B
Planner and content
needed. Mathematical Utilize the plan
Constraint conditions (after some
Optimization
Machine adjustment and
Engine
learning confirmation if
Plan necessary)
(historical)
data Expert’s planning
patterns

Instructions
Planner

High-quality plans also make production more efficient, supply its HITSODAS system for making strategic use
reduce roll wear and the amount of warm-up material, of data collected from plants to support their operational
improve product quality, and facilitate reliable operation. maintenance.
Future plans include incorporating MLCP into OT con- HITSODAS is made up of a core set of functions
trol systems to enhance their functionality. for data collection and graphical display (HITSODAS-
BASE) together with three expansion functions: HITSO-
3. Hitachi Self-organized Diagnosis DAS-QA (Quality Analysis), HITSODAS-DS (Dynamic
and Analysis System (HITSODAS) Synergy Control), HITSODAS-PH (Process Human
Interface). Figure 3 shows a block diagram of the system
The requirements for improving the quality and yield of flat and its data flow.
steel products include reducing the length of off-gauge strip, Using HITSODAS to work with data enables effec-
preventing strip breakage, and maintaining consistent strip tive system improvements to be made through an iterative
flatness. While in-depth analysis is needed to achieve these cycle in which the user extracts data from the data field
objectives, the problem with this is that steel is produced (sense), considers ways of using this data to solve problems
in a wide range of grades and strip thicknesses and widths, (think), and then makes improvements to the actual control
such that not only is the manual analysis of large amounts of systems (act).
time-series data very time-consuming, the results also tend to The “sense” step involves extracting and presenting plant
vary depending on the skill of the person doing the analysis. data from the data field. The data field is supplied with
To address these challenges, Hitachi has drawn on the plant data from the HITSODAS-BASE data collection
extensive expertise it has built up over time to develop and system. HITSODAS-PH also gives access to video from

Figure 3 — HITSODAS Data Flow

HITSODAS is equipped with functions for
• Feature analysis of plant
improving the quality of steel products Think data
that provide ongoing access to plant and • Detection of potential
equipment data. anomalies
HITSODAS-QA

• Data collection and • Automatic tuning of

Sense presentation Act control models
• Synchronized playback
HITSODAS-BASE
HITSODAS-DS
HITSODAS-PH

Data field

Plant equipment

QA: Quality Analysis DS: Dynamic Synergy Control PH: Process Human Interface

102.
 FEATURED ARTICLES

plant cameras that can be replayed in sync with the data to these changes also manifest in the motor current. By
provide a broader view of what is happening in the plant. extracting feature values from these tiny changes in motor
The “think” step involves using the data analysis and current behavior that are indicative of the degradation in
other tools provided by HITSODAS-QA to look for the the system and using a machine learning technique (a form
causes of problems and their solutions in the collected of AI developed by Hitachi), highly precise predictive
data. Detecting signs of abnormal behavior in the signals fault detection can be achieved by identifying motor
collected from the plant can help to quickly track down abnormalities in the available data (see Figure 4).
where the causes of product issues lie. The main features of the solution are as follows.
The “act” step involves conducting testing on the control (1) Greater workplace safety as inspection and diagnosis
systems based on the potential solutions identified in the can be performed remotely
“think” step. HITSODAS-DS provides functions that (2) Lower maintenance workload and costs achieved by
make the task of using plant data for control model tuning consolidating the monitoring, analysis, and diagnosis of
more efficient. multiple motors
Hitachi launched the latest version of HITSODAS (3) Ability to individually prioritize motor maintenance as
featuring enhanced graphical functions and ease-of-use the condition of each motor is monitored separately
in 2019. The user base is growing, especially outside Japan, (4) Help to address the shortage of experienced staff by
including for the operational support service that uses the standardizing work that previously relied on worker
software. intuition and rules of thumb
(5) Facilitation of highly sensitive diagnosis by extracting
4. Efficient Maintenance feature values on the basis of engineering knowledge from
motor manufacturers
Using Predictive Diagnosis
System installation is simplified by the use of clamp
of Motor Currents
sensors that fit over cables to measure the currents needed
Unplanned equipment outages or higher maintenance and for predictive diagnosis, with everything housed together
inspection costs are among the issues that arise with the in a control panel in the electrical room. This reduces
electric-motor-driven systems used at a wide variety of the number of site visits needed for sensor maintenance
industrial plants. To address these issues, Hitachi has launched compared to the past practice of attaching the sensors to
a predictive diagnosis solution that uses AI to analyze motor the motors themselves.
currents and automatically detect and highlight problems The current data is analyzed by the AI-based predictive
such as motor degradation or abnormal operation. diagnosis algorithm to indicate where signs of motor
The predictive diagnosis technology used by the solution degradation are present, providing data about the extent
draws on Hitachi’s experience with motor manufacturing of degradation for each motor. By only performing
and maintenance, incorporating OT knowledge and maintenance on those motors in need of repair or
Hitachi’s own AI analysis techniques. As degradation in replacement, this helps with both preventing unexpected
motor-driven systems causes changes in the motor load, equipment outages and reducing maintenance workloads.

Figure 4 — Overview of Motor

Predictive Diagnosis Solution
Collection of motor current data This solution uses Hitachi’s patented
Problem predictive diagnosis algorithm to analyze
motor current data and detect abnormal
motor behavior.

Gearbox Motor Control panel Current sensor

Load Current data*
(machinery, etc.)

Motor A Deterioration
abnormality
Degree of

Analyze motor currents to Problem

highlight degradation or After
other abnormalities repair

Time
Maintenance staff
Motor Predictive Diagnosis Solution

* The means by which current data is acquired differs depending on the equipment being driven.

Hitachi Review Vol. 70, No. 2 214–215 103.

 Authors
While the service is currently restricted to alternating Tsutomu Ito
current (AC) motors only, support for direct current (DC) Automation Metals Solutions Department, Industrial
motors is being added along with the development of a Manufacturing Solutions Department, Industrial
Solutions & Services Business Division, Industry &
more advanced version of the solution that can also visualize Distribution Business Unit, Hitachi, Ltd. Current work
and diagnose the condition of the motor gearbox and the and research: Development of digital solutions for
metal industry.
downstream equipment it is being used to drive.
Along with reducing staff workloads, the service also Zhe Cao
makes maintenance more efficient and supports safe and Industrial IT Solution Department, Industrial
Manufacturing Solutions Department, Industrial
reliable plant operation. Solutions & Services Business Division, Industry &
Distribution Business Unit, Hitachi, Ltd. Current work
and research: Development of digital solutions.
5. Conclusions
Kenji Maegawa
This article has described some of the solutions that Hitachi Metal Control Systems Engineering Department,
Water and Metal and Industrial Control Systems
can offer to enhance the steel production value chain. In Division, Control System Platform Division, Services
the future, Hitachi intends to continue delivering value to & Platforms Business Unit, Hitachi, Ltd. Current work
customers by offering solutions that draw on its compre- and research: Development of electromechanical
control systems for metal industry.
hensive capabilities, combining its expertise in OT and IT
to address the ever-changing challenges facing customers. Naoya Eimaeda
Automation Metals Solutions Department, Industrial
Manufacturing Solutions Department, Industrial
Solutions & Services Business Division, Industry &
Distribution Business Unit, Hitachi, Ltd. Current work
References and research: Development of digital solution for
metal industry.
1) M. Kayama et al., “Analysis System for Steel Manufacturing
Plant: How to Handle Plant Data for their Strategic use to Taku Saito
Control and Diagnosis,” The Papers of Technical Meeting on Smart Industry Department, Business Development
Metal Industries Division (MID), IEE Japan, Vol. 22, pp. 7–11 Division, Business Incubation and Development
(Oct. 1999) in Japanese. Division, Social Innovation Business Division, Hitachi,
2) K. Kuribayashi et al., “Advances in Steel Industry Control Ltd. Current work and research: Development of
Systems in Era of IoT,” Hitachi Review, 65, No. 5, pp. 41–46 digital solution for metal industry.
(Jun. 2016).

104.