Blog

Why Industrial Improvement Fails When It Starts with Technology

Industrial improvement often fails when organizations choose technology before defining the underlying engineering problem. Dashboards, historians, AI tools, and automation systems add value only when they match plant behavior, process constraints, equipment reliability, and the decisions operators and managers need to make.

The post argues for starting with first-principles engineering: identify the constraint or failure mode, understand the physical and operational causes, define the required action, and then select the simplest technology that supports it. It concludes that practical, measurable improvement begins with understanding rather than tools.

Industrial improvement often fails because it starts with a technology answer before the engineering question has been properly defined. A new dashboard, historian, PLC platform, AI tool or automation system may look like progress, but unless it is connected to the real behaviour of the plant, the constraints of the process, the reliability of the equipment and the decisions people need to make, it becomes another layer of complexity.

1. Introduction — The common mistake

Many industrial improvement programmes begin with a technology choice: a platform, dashboard, historian, AI tool, SCADA upgrade, cloud system, sensor network or automation package.

That is often the wrong starting point.

The correct starting point is the plant and the business problem.

2. The real industrial system

Industrial performance is shaped by several interacting layers:

  • electrical power supply;
  • equipment condition;
  • control systems;
  • instrumentation;
  • operator behaviour;
  • maintenance strategy;
  • production constraints;
  • data quality;
  • business decision cycles.

When these are not understood together, technology tends to add complexity rather than value.

3. Why data alone does not create insight

A historian, dashboard or analytics tool does not automatically improve production. If the plant data is incomplete, poorly structured, badly contextualised or not connected to decisions, it becomes another system people ignore.

Useful data must answer operational questions.

Examples:

  • Is the plant stable?
  • What is constraining production?
  • What failure mode is developing?
  • What risk is increasing?
  • What decision must be made now?
  • What action will improve safety, reliability or throughput?

4. Why automation alone does not create productivity

Automation can make a good process more repeatable. It can also make a poorly understood process fail faster.

Effective automation requires:

  • sound process understanding;
  • reliable instrumentation;
  • stable electrical and control systems;
  • clear operating philosophy;
  • maintainable logic;
  • appropriate alarms;
  • practical operator interfaces;
  • proper commissioning and handover.

5. The role of first-principles engineering

Before selecting technology, define:

  • What problem are we solving?
  • What is the current failure or constraint?
  • What physical, electrical or process mechanism is involved?
  • What information is needed?
  • What action will the system or person take?
  • How will success be measured?
  • What new risks are introduced?

This is where engineering judgement matters.

6. A practical framework

A useful industrial improvement approach could be:

  1. Understand the plant and production objective.
  2. Identify the constraint, risk or failure mode.
  3. Confirm the physical and operational causes.
  4. Assess electrical, control, instrumentation and maintenance dependencies.
  5. Define the required decision or control action.
  6. Select the simplest technology that supports that action.
  7. Commission, verify, document and train.
  8. Measure the result and adjust.

7. Conclusion

Industry 4.0, automation, AI, analytics and digital transformation all have value. But they only create value when grounded in industrial reality.

The best improvements do not begin with technology.

They begin with understanding.

The useful question is not “What technology should we install?” The useful question is “What must we understand, control or decide better than we do today?” Once that is clear, technology can be selected and engineered to serve the plant. That is where industrial improvement becomes practical, measurable and sustainable.


Integrated Management Systems: From Operational Firefighting to Effective Management

F. Grant Schutte’s Integrated Management Systems, published in 1981, presents a remarkably enduring framework for understanding how organisations should distribute management effort.

At the heart of the model is a distinction that remains highly relevant:

Effective management is concerned with doing the right things. Efficient management is concerned with doing things right.

An organisation requires both.

The difficulty is that managers frequently become absorbed in operational problems at the expense of the planning and control activities appropriate to their position. Senior managers become involved in matters that should have been delegated, operational information flows unnecessarily upwards through the organisation, and long-term planning is displaced by immediate problems.

Schutte describes the desired state as the Normative Management Structure and contrasts it with the frequently observed Empirical Management Structure.

The difference between the two provides a useful way of examining how organisations are actually managed.

Three Levels of Planning and Control

The management system can be divided into three broad activities:

  1. Strategic Planning
  2. Management Planning and Control
  3. Operational Control

These activities are related, but their objectives, time horizons, information requirements and appropriate management levels are fundamentally different.

Strategic Planning

Strategic planning establishes where the organisation is going.

It is concerned with:

  • defining the mission of the organisation;
  • determining the businesses, products or markets in which it intends to participate;
  • establishing organisational objectives and changes to those objectives;
  • allocating resources; and
  • establishing strategic guidelines and policies that provide direction to subsequent management planning.

Strategic planning therefore addresses questions such as:

What business are we in?

What should the organisation be trying to achieve?

Where should resources be committed?

What strategic principles should guide management decisions?

This activity belongs primarily to top management.

Its time horizon tends to be long, the problems are frequently unstructured, and the information available is often uncertain and forward-looking. Strategic decisions consequently require substantial analytical and creative judgement.

An effective organisation must also anticipate change rather than merely react to it. Strategic management should therefore develop competitive strategies before they are required, determine appropriate products and markets, consider sources of raw materials and resources, apply innovation to organisational processes, and develop the human capabilities required for future productivity.

This is what Schutte characterises as doing the right things.

Management Planning and Control

Management planning and control occupies the middle ground between strategy and day-to-day operations.

Its purpose is to ensure that resources are obtained and used effectively in achieving the organisation’s objectives.

This includes:

  • developing plans for acquiring resources;
  • developing programmes for using those resources;
  • allocating resources between competing organisational requirements;
  • monitoring performance against plans; and
  • replanning when circumstances or performance require it.

The emphasis here is particularly important.

Management control should not simply consist of reporting historical performance.

Its purpose is to trigger replanning.

A useful management control system therefore identifies meaningful departures from plan and causes management to reconsider actions, resources or assumptions.

In this sense, management control is dynamic rather than passive.

It is principally the responsibility of line and middle management, although top management remains significantly involved.

Operational Control

Operational control concerns the efficient execution of individual tasks and transactions.

It requires:

  • clearly specifying the task;
  • establishing procedures for performing it;
  • defining acceptable standards;
  • measuring actual performance;
  • identifying deviations from those standards; and
  • initiating corrective action.

Operational control therefore asks:

Was the task performed correctly?

The focus is comparatively short term and highly structured. Information is normally accurate, detailed and historical. Procedures, rules and standards play a much greater role than they do in strategic decision-making.

Examples include production output, equipment performance, quality, cost, maintenance execution, process compliance and other measurable operating activities.

This is the domain of doing things right.

Effectiveness and Efficiency Are Different Management Problems

A well-managed organisation needs both effectiveness and efficiency.

Effectiveness arises predominantly from:

Strategic Planning → Management Planning and Control

Efficiency arises predominantly from:

Operational Control

This distinction matters.

An organisation can be extremely efficient while pursuing the wrong objectives. It may manufacture an unwanted product exceptionally well, reduce the cost of an obsolete process, or optimise operations within a fundamentally poor business strategy.

Conversely, an organisation can choose an excellent strategy but fail because its operational execution is poor.

Successful management therefore requires the integration of both.

The Normative Management Structure

Schutte’s normative model considers how management attention should ideally be distributed through an organisation.

The normative management structure illustrates the degree of effort expended on management activities at each management level.

Top Management

Top management should be heavily involved in:

  • strategic planning; and
  • management planning and control.

Top management should become directly involved in operational control only in exceptional circumstances.

Middle Management

Middle management should be heavily involved in:

  • management planning and control; and
  • operational control.

It should also contribute to strategic planning, although to a lesser degree than senior management.

Operational Management

Operational or supervisory management should concentrate primarily on:

  • operational control.

It should participate to a smaller extent in management planning and control and generally should not be responsible for strategic planning.

Importantly, these proportions concern the management time devoted to planning and control activities, rather than the manager’s total working time.

The model therefore establishes a logical division of managerial attention.

The higher the managerial position, the greater the emphasis should become on effectiveness and the future.

The lower the position, the greater the emphasis should become on efficient execution of present operations.

Characteristics of the three management horizons: Strategic Planning, Management Planning and Control, and Operational Control.

What Happens in Practice?

Real organisations frequently depart significantly from this model.

Schutte describes this condition as the Empirical Management Structure.

Managers at progressively higher levels become drawn into operational control.

Instead of concentrating on planning, resource allocation and management control, senior managers become occupied with detailed operating problems.

The result is familiar in many organisations:

  • excessive management attention is directed towards immediate problems;
  • insufficient innovative or strategic planning occurs;
  • senior people solve problems that should have been resolved lower in the organisation;
  • operational information is reported to managers who do not need it;
  • decision-making becomes unnecessarily centralised; and
  • management becomes reactive rather than anticipatory.

The organisation may appear extremely busy while still being poorly managed.

Empirical management structure showing the downward displacement of managerial effort into lower-level control activities, reducing attention to strategic planning and effective business control.

Why Managers Become Trapped in Operational Control

Several causes contribute to the empirical structure.

A manager may be uncertain about the real purpose of their position.

Large quantities of operational information may flow upward, creating pressure to become involved simply because the information is available.

Formal management control and replanning systems may be inadequate, making it difficult to manage by exception.

A technical specialist promoted into general management may continue behaving primarily as a specialist.

Managers may also simply fail to delegate.

One particularly recognisable phenomenon occurs when a manager is promoted but effectively takes the previous job with them.

Instead of allowing the subordinate manager to assume responsibility, the promoted manager continues performing significant parts of the old role.

The hierarchy then becomes compressed: several management levels become involved in essentially the same operational decisions while higher-level management responsibilities receive insufficient attention.

Delegation Is More Than Assigning Work

The ability to establish the normative management structure depends heavily on proper delegation.

Delegation requires three things.

First, the subordinate must be told clearly what is expected. Objectives, standards and procedures must be sufficiently defined to guide action and permit subsequent evaluation.

Second, the subordinate must have the knowledge and capability to perform the responsibility. Training and development may therefore be necessary.

Third, the subordinate must receive the resources and authority required to perform effectively and efficiently.

Delegating responsibility without delegating sufficient authority is not genuine delegation.

Neither is assigning work without establishing the expected result.

Specificity and Accountability

Schutte identifies two particularly important principles underlying delegation.

The Principle of Specificity

It must be clear exactly what has been delegated.

Everyone should know:

Who is responsible for what?

Without clearly defined responsibilities, performance cannot be meaningfully controlled or evaluated.

Ambiguous responsibility also creates one of the classic organisational problems: several people appear responsible when things go well, while nobody is responsible when things go wrong.

The Principle of Accountability

Delegating an activity does not eliminate the delegating manager’s accountability.

A manager remains accountable for the responsibilities delegated to subordinates and must therefore exercise appropriate management control over them.

This does not mean continually taking the work back.

Instead, the manager participates at the appropriate level in analysing problems, evaluating opportunities and making decisions associated with the responsibilities that have been delegated.

This leads to an important distinction:

Delegation transfers responsibility for performing the work, but it does not eliminate managerial accountability for its management.

Key Performance Areas

The mechanism used to define managerial responsibility within the system is the Key Performance Area, or KPA.

A KPA is a managerial result that is particularly important to the performance of the organisation.

The concept applies Pareto’s principle to management: a relatively small number of results typically make a disproportionately large contribution to organisational performance.

Consequently, a manager’s most important responsibilities are not necessarily the activities that consume most of their time.

A managerial position would normally have only a small number of meaningful KPAs — typically three to five.

This forces attention onto what actually matters.

Key Performance Areas, associated performance measures, and the structured flow and frequency of management control reporting through the organisational hierarchy.

What Makes a Good KPA?

A properly constructed KPA has several important characteristics.

It should be:

  • unique to the particular position within that organisational branch;
  • expressed as an end result, rather than an activity;
  • measurable, preferably objectively;
  • aligned with the KPAs of other positions;
  • relevant to organisational performance; and
  • consistent with management’s strategy for operating the organisation.

A KPA should therefore not simply say:

“Prepare production reports.”

That describes an activity.

A result-based KPA might instead concern:

Production performance against agreed quantity, quality and cost targets.

The distinction changes the management focus from performing activities to achieving results.

The Common KPA of Every Manager

Although individual positions have unique KPAs, managers also share an important common responsibility:

Managing the KPAs of their subordinates.

This follows directly from the principle of accountability.

A manager who delegates work must establish an appropriate mechanism for determining whether the delegated results are being achieved.

However, this does not require every manager in the hierarchy to receive the same detailed information.

Quite the opposite.

Management Information Should Become More Summarised as It Moves Upward

One of the useful consequences of a KPA-based management system is that information can be designed around management need.

Operational management may need detailed daily information.

Higher levels may require only summaries or information about significant exceptions.

For example, production management at operating level might require detailed daily information concerning:

  • quantity;
  • quality; and
  • cost.

A superintendent may require similar information but increasingly in summarised form.

Senior management may need only periodic summaries indicating whether the KPA is under control and whether intervention or replanning is required.

The objective is therefore not to maximise information.

It is to provide the right information, at the right level, at the right frequency, for the appropriate management decision.

Eliminating redundant control information is consequently an important element in moving from empirical to normative management.

Hierarchy of Key Performance Areas and the corresponding yardsticks used to measure effectiveness and efficiency across the management horizons.

Developing KPAs Through Participation

Schutte proposes a participatory process for identifying KPAs.

The first step is to identify the managerial structure and select an analysis group.

This would normally include:

  • the manager whose position is being analysed;
  • managers above that position through the relevant chain of command; and
  • peers with whom the manager interacts.

The analysis then examines the activities currently performed by the manager.

Existing job descriptions and practical experience can be used to develop an agreed list.

The critical next step is to translate those activities into outputs or end results.

Pareto analysis can then be applied to identify which outputs genuinely represent Key Performance Areas.

Useful questions include:

Does this result materially contribute to organisational performance?

and:

If the manager performed poorly in this area, could that person still reasonably be regarded as performing the managerial role successfully?

For every KPA, the organisation should then determine:

  • the performance yardstick;
  • the information required to measure it;
  • how each management level intends to exercise control; and
  • how frequently information should be reported.

The result is not merely an improved job description.

It is the foundation of a management control system.

From Empirical Management to Normative Management

The ultimate objective is to move the organisation away from unnecessary senior-management involvement in operational control.

The transition requires:

Clear organisational responsibilities

Well-defined KPAs

Effective delegation

Appropriate performance measures

Purpose-designed management information

Management by exception

Dynamic replanning

Greater managerial attention to effectiveness

When management information is structured around KPAs, managers no longer need to continuously inspect large volumes of operational detail.

Performance outside agreed limits becomes the trigger for management attention.

Management control therefore becomes a mechanism for deciding when replanning or intervention is necessary, rather than a mechanism for encouraging continual interference in subordinate activities.

A Management Principle That Remains Relevant

Although Integrated Management Systems was published in 1981, the underlying management problem has hardly disappeared.

Modern organisations possess vastly more operational information than the organisations of that period.

Dashboards, ERP systems, live production data, maintenance systems, financial reporting platforms and automated notifications can place extraordinary quantities of information in front of managers.

That capability does not necessarily improve management.

Without clearly defined responsibilities and KPAs, better information systems can simply make it easier for senior managers to become involved in matters that should be controlled elsewhere.

The critical question is therefore not:

“What information can we give this manager?”

It is:

“What decisions is this manager accountable for making, and what information is necessary to make those decisions?”

That distinction lies at the heart of an integrated management system.

An effective management structure allows operational managers to control operations, middle managers to manage resources and performance, and senior managers to concentrate on direction, strategy and the future.

The objective is not to remove control.

It is to place control at the correct level.

And perhaps the most useful test of any management structure is therefore a simple one:

Are our managers spending their time managing the responsibilities appropriate to their position — or are they still doing the work of the people below them?


Developing a Continuous Caster Breakout Detector: An Engineering Case Study

In the early 1990s, the continuous casting plant at Iscor’s Pretoria Works faced a serious operational problem: sticker breakouts.

A breakout in a continuous caster is not merely a production interruption. It involves molten steel escaping from the partially solidified strand, with the potential for extensive equipment damage, prolonged production downtime and significant risk to personnel.

By 1993, the Pretoria Works caster was experiencing approximately two to three sticker breakouts per month. Rather than purchase a proprietary commercial detection system, a small multidisciplinary engineering team set out to understand the phenomenon and develop an in-house solution.

The result was a continuous caster breakout detector combining mould temperature measurement, supervisory control, mathematical modelling and an expert-system decision engine.

It became an early example of what would now be described as industrial condition monitoring combined with real-time predictive control.

What Is a Sticker Breakout?

During continuous casting, molten steel at approximately 1540 °C is poured into a water-cooled copper mould.

As the steel contacts the mould walls, its outer surface begins to solidify. A solid shell therefore develops around a still-molten core while the strand is continuously withdrawn from the mould at approximately 1 metre per minute.

Under normal operating conditions, mould powder provides lubrication between the solidifying strand and the copper mould.

If lubrication becomes inadequate, friction may increase sufficiently for part of the solidifying steel shell to stick to the mould wall near the meniscus.

The strand, however, continues moving downwards.

The result is a tear in the newly formed shell.

Because this occurs beneath the surface of the molten steel, the operator cannot see it. As withdrawal continues, the tear propagates downward through the mould. Unless the shell can heal before reaching the mould exit, molten steel escapes from the strand.

This is the sticker breakout.

The consequences can include damage to the mould and downstream segments, lost steel, substantial repair effort and several hours of production downtime.

The mechanism of a sticker breakout showing the progression of temperature associated with slab wall thickness.
The scar of a prevented breakout on the continuously cast steel slab.

The Engineering Problem

At the time, Pretoria Works had no practical means of predicting or detecting a developing sticker breakout.

Commercial detection systems were available, but their installed cost was estimated at between R2 million and R3 million. They also presented another engineering disadvantage: their proprietary nature made them difficult to understand, modify and optimise for local casting conditions.

For the project team, this presented an opportunity.

Could the plant develop a system internally that was:

  • effective at detecting developing sticker breakouts;
  • fast enough to allow corrective action;
  • inexpensive compared with proprietary alternatives;
  • based on standard, maintainable hardware;
  • understandable by plant engineering and maintenance personnel;
  • adaptable to the particular behaviour of the Pretoria caster; and
  • capable of improving the plant’s understanding of the continuous casting process itself?

A preliminary investigation was therefore approved.

The ultimate objective was ambitious: reduce the breakout frequency to less than one per year.

Detecting a Breakout Through Temperature

The key to detecting a sticker breakout lies in understanding what happens thermally when the steel shell sticks to the mould.

Thermocouples installed in the mould allow the temperature distribution across its faces to be monitored continuously.

A developing sticker produces a characteristic change in this temperature pattern.

Rather than relying on a single temperature limit, the project sought to identify the spatial and temporal temperature behaviour associated with a developing tear.

This required much more than installing instrumentation.

Temperature histories had to be gathered during actual casting operations, archived and compared with plant events. Literature describing the theoretical breakout mechanism was then compared with what was actually observed on the caster.

Over time, recognisable patterns emerged.

The engineering challenge was then to convert those patterns into mathematical rules capable of distinguishing a genuine developing sticker from the many normal temperature variations occurring during casting.

From Plant Data to an Expert System

Development proceeded along two parallel paths.

The first concentrated on refining the physical measurement system.

The second concentrated on developing a software decision-support system capable of evaluating the measured temperatures and determining whether a sticker breakout was developing.

The system ultimately combined:

  • mould-mounted thermocouples;
  • signal acquisition hardware;
  • personal computers communicating over a NetBIOS network;
  • commercial supervisory control and data acquisition software;
  • a proprietary decision-support application written in C++;
  • mathematical breakout-prediction algorithms;
  • operator displays; and
  • integration with the plant programmable logic controller.

For the early 1990s, this was an unusually distributed and software-intensive industrial control application.

Importantly, most of the system was constructed from commercially available, off-the-shelf components.

Only the specialised temperature-sensing arrangement and the breakout decision software were proprietary developments.

This deliberately avoided the “black box” problem associated with proprietary commercial systems. Plant engineers could understand how the system worked, maintenance personnel could replace standard components, and the detection parameters could be modified if operating conditions changed.

Components of the Breakout Detector.

Engineering Was Only Part of the Project

A technically correct detector would have achieved little without the cooperation of the people operating and maintaining the caster.

The core project team consisted of three people:

Alf Jansen van Rensburg — project leader, Process Control
Nico Davies — hardware, supervisory control and data acquisition
Christelle du Raan — development of the expert-system decision logic

However, the effective project team was considerably larger.

Maintenance personnel modified the mould, installed instrumentation and supported thermocouple replacement.

Production personnel contributed their operating experience.

This proved particularly important.

Operators had accumulated considerable practical knowledge concerning mould lubrication, casting behaviour, mould powder and abnormal operating conditions. Their anecdotal observations helped the engineering team interpret temperature behaviour that could not be understood from theory alone.

The development process therefore became a combination of:

theoretical knowledge + measured plant data + operator experience

That combination was fundamental to the performance of the finished system.

Human-Machine Interface Design

The operators were also actively involved in determining how information should be presented.

Different representations of mould temperature behaviour were evaluated, with operating personnel identifying those that best matched their understanding of the casting process.

Control-room operators, shift personnel and engineers were trained to interpret the system.

Mould operators were trained in the actions required following a breakout warning.

Initially, the principal response was to reduce strand withdrawal speed, providing additional time for the torn shell to heal.

As confidence in the detector increased and false alarms became sufficiently rare, the system progressed from decision support to automatic intervention.

When a sticker was detected, the casting machine could be stopped automatically through the plant PLC, although the operator retained a short opportunity to override the automatic action.

This represented an important transition.

The system was no longer simply monitoring the process.

It had become part of the process control strategy.

Measuring the Results

The operational results were substantial.

Before installation of the detector, the Pretoria Works continuous caster averaged approximately:

2 sticker breakouts per month

Immediately following commissioning, this reduced to approximately:

1 breakout every 2 months

The breakout frequency had therefore fallen to approximately one quarter of its previous level.

The chart shows a dramatic change in the frequency of breakouts after commissioning the Breakout Detector.

The project report calculated the average cost of a breakout at R359,379, including production downtime, labour, equipment damage and lost steel.

Development of the complete system, including engineering labour and purchased equipment, was calculated at approximately:

R1.025 million

The cost of the system was consequently recovered through approximately the first three prevented breakouts.

Those prevented breakouts occurred within approximately the first two months of operation.

The annual saving resulting from reducing the breakout rate from two per month to 0.5 per month was conservatively calculated at approximately:

R6.47 million per year

The corresponding increase in steel slab production was calculated at approximately:

7,752 tonnes per year.

At the target production rate of 45,000 tonnes per month, the project was estimated to reduce the variable production cost of steel slab by approximately 1.92%.
All monetary values quoted above are historical South African rand values from the original 1996 project report and should not be interpreted as present-day values.

Benefits Beyond Production

The financial return was significant, but the system produced several other benefits.

Improved Safety

A breakout releases molten steel into the caster structure.

Preventing the event therefore reduced the exposure of production and maintenance personnel to both the breakout itself and the difficult cleanup and repair activities that followed.

Reduced Maintenance Work

Preventing damage to the mould and caster segments reduced emergency maintenance work.

Workshop capacity previously consumed repairing breakout damage could instead be directed towards preventive maintenance and other productive work.

Improved Process Information

The temperature monitoring system gave production personnel far greater visibility into conditions within the mould.

Among other things, the temperature distributions could provide information regarding the effectiveness of mould lubrication.

This allowed operating personnel to respond to inadequate lubrication before it developed into a more serious condition.

Improved Slab Quality

Better mould lubrication does not merely reduce the probability of sticking.

It also contributes to improved slab surface quality.

The breakout detector therefore became useful not only as a protective system, but also as a process diagnostic instrument.

Knowledge Retention

Because the system was internally developed and deliberately constructed from understandable components, the knowledge remained within the organisation.

Detailed documentation and training material were developed for production, maintenance and engineering personnel.

The system therefore became part of the plant’s accumulated technical knowledge rather than remaining a proprietary technology understood only by an external supplier.

Designing for Maintainability

Maintainability was an explicit design objective from the beginning.

The thermocouples were standard commercial components.

The computer and data-acquisition hardware consisted largely of standard products.

Detection parameters were accessible through the expert-system interface, subject to password protection.

The C++ source code was retained so that future engineers could modify the application if required.

This approach proved successful.

The completed breakout detector was sufficiently stable that its maintenance burden was extremely small, apart from occasional replacement of mould thermocouples.

That design philosophy remains relevant today.

Industrial systems often remain in service far longer than the computers, software packages and communications technologies originally used to construct them.

A system built around open interfaces, understandable logic and replaceable components has a much better chance of remaining supportable throughout its operational life.

Lessons From the Project

Looking back three decades later, several aspects of the project remain particularly relevant to modern industrial automation.

1. Understand the Process Before Automating It

The project did not begin with software.

It began by understanding the physical mechanism that produced the failure.

Only once the thermal behaviour of a developing sticker was understood could meaningful detection logic be developed.

2. Plant Data Must Be Interpreted in Context

Raw temperature data was not enough.

The theoretical models had to be reconciled with actual plant behaviour.

Production knowledge was essential to explaining deviations from theory.

3. Operators Are Part of the Control System

The people operating the caster were not treated merely as end users.

Their knowledge influenced the detection model, the interface and the operating response.

That involvement was important in establishing confidence in the system.

4. Transparency Has Engineering Value

The decision to avoid a proprietary black-box solution provided benefits extending well beyond initial capital cost.

It allowed the plant to understand the technology, modify it and use it as a platform for further process investigation.

5. Automation Should Be Introduced According to Confidence

The system initially provided warnings and operator guidance.

Only after sufficient operating experience demonstrated reliable detection and a sufficiently low false-alarm rate was automatic machine intervention introduced.

This remains a sound philosophy for implementing predictive control systems.

6. The Greatest Value May Be the Knowledge Created

The detector prevented costly failures, but it also created a new capability.

Continuous mould temperature measurement provided metallurgical and process engineers with a platform for studying the influence of casting speed, mould oscillation, mould taper, mould powder, superheat and other parameters on caster behaviour.

The project therefore evolved from solving one operational problem into providing a broader process-development tool.

From Expert Systems to Modern Predictive Analytics

The terminology used today would be different.

A similar project might now be described using terms such as:

  • edge computing;
  • predictive analytics;
  • anomaly detection;
  • multivariable condition monitoring;
  • machine learning;
  • digital twins; or
  • industrial artificial intelligence.

But the fundamental engineering problem has not changed.

Measurements must reliably represent the physical process.

The failure mechanism must be understood.

The distinction between normal process variation and an incipient failure must be identified.

The detection system must respond quickly enough to matter.

And operators must trust the result.

The Continuous Caster Breakout Detector developed at Pretoria Works achieved these objectives using the technology available in 1993–1994.

Perhaps the most important lesson from the project is therefore not about any particular technology.

It is that deep process understanding, good instrumentation, disciplined analysis and close cooperation between engineering, operations and maintenance can solve problems that initially appear to require expensive proprietary solutions.

That principle remains as valid today as it was thirty years ago.

Pretoria Works Focus Newsletter


Industrial Implementation of a Fundamental Model for Electric Arc Furnace Control

From process theory to a working industrial control solution

Electric Arc Furnace (EAF) operation is a highly dynamic metallurgical process. Scrap and direct reduced iron are melted while electrical energy, oxygen, carbon, lime and dolomite are added continuously, and the operating conditions change throughout the heat.

The challenge addressed by this project was to move beyond control based primarily on fixed operating practice and operator experience, and instead use a fundamental process model to estimate the current state of the furnace, predict its future behaviour and calculate appropriate control actions.

The objective was not simply to develop a mathematical model. The real engineering challenge was to implement that model in an operating steel plant and make it sufficiently robust, practical and understandable to become part of normal production.

The resulting system demonstrated that a detailed fundamental model could successfully control an industrial EAF despite significant variations in raw-material quality and process conditions.

Author’s perspective

Looking back after many years, I still carry the scars of a highly technical and demanding project. We were implementing technology years ahead of its time, while also managing team morale, project delivery and a frequently adversarial client environment.

It remains one of the most satisfying projects I have ever delivered. I am enormously proud of Christelle for developing the fundamental model from the ground up and then seeing it realised successfully in practice.

The team persevered. In the end, both the Client and the Engineering House achieved their objectives: proven process control that efficiently delivered steel to specification at an optimised cost.

The process challenge

An electric arc furnace is particularly difficult to model and control because many process variables interact simultaneously.

The process involves electrical energy input through the furnace electrodes together with chemical energy from oxygen and carbon injection. Lime and dolomite additions influence slag chemistry, while scrap and direct reduced iron feed continually change the thermal and metallurgical condition of the furnace.

Many important process states cannot be measured continuously. Operators therefore traditionally infer furnace conditions from a relatively small number of available measurements and from experience accumulated over many heats.

This creates a difficult control problem: the system is highly nonlinear and dynamic, yet many of the most important internal states are only indirectly observable.

The project addressed this by constructing a model based on the fundamental mass, energy and metallurgical relationships governing the EAF process. The model could then estimate conditions that could not be measured directly and use those estimates to determine appropriate process control actions.

A model designed around the actual furnace process

The control system was structured around a series of calculation functions rather than a single monolithic model.

These included:

  • offline heat-recipe calculations;
  • static model calculations;
  • online recalculation and prediction;
  • sample-based adjustment and recalculation;
  • dynamic model set-point adjustment; and
  • production-route and alloy-addition calculations.

Together, these functions allowed the model to operate throughout the complete furnace heat rather than only at isolated operating points.The furnace heat itself was represented as a sequence of operating stages. Typical stages included preparation, melting of scrap baskets, refining and bulk DRI addition, followed by final adjustment and tapping.

For every stage, the process model calculated the material additions and operating set points required to move the furnace towards its required final state.

This was important because the model was not simply calculating an end-point target. It was continually determining how the furnace should be operated to reach that target.

Closing the loop with real plant measurements

A major strength of the implementation was that the model did not assume that calculated conditions would exactly match the real furnace.

At the start of a heat, the model worked from the available charge information and recent process measurements. As the heat progressed, the calculation was repeatedly updated using actual plant measurements.

Differences between predicted and measured conditions were therefore fed back into subsequent calculations.

The model could account for changes in factors such as:

  • material feed;
  • actual electrical power transfer;
  • furnace pressure;
  • slag-door condition;
  • oxygen and carbon additions; and
  • measured temperature and sample results.

This produced a continually revised estimate of the furnace condition and provided the basis for calculating the remainder of the heat.

This adaptive behaviour was essential in an industrial environment because real raw materials and operating conditions never exactly match the assumptions used during initial recipe calculation.

Dynamic optimisation during the heat

The model was also designed to optimise important operating variables as the process progressed.

For example, during refining and DRI addition the model could determine suitable profiles for:

  • DRI feed rate;
  • roof carbon addition;
  • oxygen flow;
  • lime and dolomite requirements; and
  • energy input.

The purpose was to continuously move the furnace towards the required temperature, composition and slag condition rather than rely solely on fixed operator set points.

The model therefore became part of the operating strategy of the furnace rather than simply functioning as an advisory calculation tool.

Demonstrating control on the operating furnace

The most important test of any industrial process model is not whether it reproduces historical data, but whether it can successfully control the real process.

The operating results demonstrated this capability.

During furnace operation, the model successfully controlled the process through substantial variations in scrap and DRI quality. The results also showed that the furnace could be controlled to the required operating temperature while simultaneously working towards the required chemical composition, tap mass and tapping temperature.

This is significant because these objectives are strongly coupled. Changing electrical energy, oxygen, carbon or material additions to correct one process condition can affect several others.

Successful operation therefore required the model to maintain a coherent representation of the complete furnace process.

Electric Arc Furnace kWh vs DRI Feed

The project was more than a modelling exercise

One of the more valuable lessons from the project was that implementing advanced process control is not solely a mathematical or software problem.

The paper recognised that successful implementation required a corresponding level of organisational maturity.

An advanced control system changes the way operators, metallurgists and process engineers interact with the furnace. It transfers some decisions that were historically based on operator experience into a formalised control system.

That transition requires confidence in both the technology and the organisation implementing it.

The project therefore placed considerable emphasis on defining the required functionality before implementation and on understanding how the system would interact with operators and existing plant practices.

The experience reinforced an important engineering principle:

technology implementation begins with defining what the process needs—not with selecting the technology.

Defining the solution before building it

A key part of the implementation approach was the development of a clear functional specification.

The project emphasised defining:

  • what the process model was required to achieve;
  • what information was available from the plant;
  • what control actions the model would be permitted to make;
  • how the model would interact with operators;
  • how abnormal conditions would be handled; and
  • how the resulting system would fit into normal production.

This up-front engineering effort was considered fundamental to successful construction and commissioning.

The paper also observed that attempts to reduce this early engineering effort often simply defer the cost and complexity into construction, commissioning and plant operation.

That remains highly relevant to modern automation and digitalisation projects.

Successful industrial implementation

The central achievement of the project was that the fundamental model moved beyond simulation and became an operational industrial system.

Despite the inherent complexity of the model and the practical difficulties involved in implementing it, the project demonstrated that a successful solution could be achieved in an operating production environment.

Several aspects of the project stand out as particular successes.

A fundamental process model was successfully deployed online

The system showed that detailed mass, energy and metallurgical calculations could operate sufficiently quickly and reliably to participate directly in real-time furnace operation.

The model tolerated substantial process variability

The furnace could be controlled despite significant variations in scrap and DRI quality rather than requiring tightly controlled laboratory conditions.

Multiple process objectives were controlled simultaneously

The model did not optimise a single isolated variable. It managed the interaction between energy input, material additions, temperature, chemical composition, slag conditions and final tapping requirements.

The model adapted during each heat

Plant measurements and samples were used to revise the estimated furnace state and recalculate the remaining operating strategy.

This provided resilience against inevitable differences between the calculated recipe and the real furnace.

The system became an operational tool rather than an engineering study

The project progressed through modelling, process definition, construction, commissioning and practical furnace operation.

That transition from theoretical model to production system was arguably the most important measure of success.

Engineering lessons that remain relevant

Although this project was undertaken in the context of electric arc furnace control, several lessons remain applicable to contemporary advanced automation, optimisation and digital-twin projects.

Fundamental models can provide capabilities that empirical control alone cannot. They make it possible to estimate important process states that cannot be measured continuously and to predict the consequences of proposed control actions.

Successful models must continually reconcile themselves with the real plant. Industrial processes rarely behave exactly as predicted, so online measurement and recalculation are essential.

Process knowledge is at least as important as software. The effectiveness of the solution depends on correctly representing the physical and metallurgical behaviour of the process.

Implementation engineering matters. Interfaces, operating procedures, abnormal conditions and operator interaction are not secondary details; they are part of the control system.

Organisational acceptance is critical. Advanced automation changes the relationship between operators and the process. Successful implementation requires both technical confidence and operational ownership.

Looking back

The project demonstrated something that is sometimes overlooked in discussions of advanced process control: sophisticated models only create value when they can survive contact with the real industrial process.

The success of this project was therefore not simply the development of a furnace model.

It was the successful integration of metallurgy, process engineering, automation, plant measurements and operating practice into a system capable of controlling a highly variable production process.

That remains the distinction between a modelling project and an industrial control solution.

Original technical paper

This article summarises work described in the conference paper:

Industrial Implementation of a Fundamental Model for Dynamic Control of an Electric Arc Furnace

The original paper provides further detail on the furnace process, model structure, implementation methodology and operating results.