Maximum efficiency, maximum profit - controlling your systems

TWI Bulletin, September/October 1991

Jim Foster joined The Manufacturing Consultancy at TWI in 1986 after a number of years in industry. His past experience in manufacturing engineering has covered manufacture of computer systems, motor vehicle brakes and clutches and radio communications equipment.

Since moving to TWI he has worked on application of vision to control of TIG welding, design and construction of TWI's 10kW laser beam manipulation system and introduction of integrated manufacture to the constructional steel work industry. Latterly his work has concentrated on factory communication systems.

An understanding of what engineers mean by control and the principles involved in controlling manufacturing systems is essential to survival in today's economic climate. Jim Foster proffers such an understanding using examples drawn from TWI's work in this field.

What is control?

To a layman 'control' is the term used to describe, for example, a knob which adjusts the current from a welding power supply unit (PSU). To a control engineer it is a matter of complex circuit design within the PSU and external sensory systems, which combine to ensure that the welding current corresponds to the dial markings. To a manufacturing systems engineer, control is the means whereby 'what is planned, is made to happen'.

Control should be applied equally to all resources (human and machine) used by the business. Regrettably the complexity and unpredictability of human resources make their control difficult except, perhaps, in a military environment. Human factors have been a major consideration in industry's considerable investment in mechanisation, automation and 'integrated manufacturing' systems. The level of control necessary to make possible automation of individual machines (and processes) and groups of machines (cells) is well understood and documented, e.g. Ref. ^[1] and ^[2] . Control requirements for integration of complete manufacturing systems (an area of greater potential than automation alone) are less well understood.

Principles

The precursor to design of any control system is a thorough knowledge of what the manufacturing system is required to do and how it is to achieve this. Such knowledge is usually expressed in a 'model' of the machine, process or system. These models must reflect all requirements as they become apparent; thus they must be easy to update. Control of manufacturing system is achieved by:

• Observation/measurement of what is happening ('reality');
• Comparison of reality with the model (what is required) and deciding whether corrective action is necessary;
• Determination of the type and amount of corrective action needed;
• Implementation of the necessary corrective action.

All control systems, regardless of purpose or complexity, follow these operating principles.

Control in practice

Manual operations

Consider production of a simple manual metal arc (MMA) weld, where control is by the human welder ( Fig.1). The model of the process exists as:

1. A statement of what is required giving a definition of weld quality (e.g. bead shape, penetration, porosity, spatter);
2. A statement of how the task is to be performed in the form of an approved welding procedure (e.g. preparation, current, technique).

Control of the manufacturing process is then achieved by the welder who:

1. Sees and hears what is happening at the arc/weld pool, observes the solidified weld and the settings of his equipment;
2. Compares what he sees and hears with his interpretation of the 'process model' (which includes the welding procedures), then decides whether any adjustment is necessary (with trained welders these actions are usually instinctive);
3. Determines (based on experience and the process model) what adjustments are to be made, their extent and how best to make them;
4. Implements the adjustment (i.e. within the scope of the process model, changes voltage setting, arc length, electrode attitude, welding current) manually.

This is often considered to be a 'simple' task, but in reality it is complex; the welder is required to consider, simultaneously, numerous inputs from several of his sensory systems and to determine, then make, adjustments. All this must be accomplished in 'real time' (i.e. sufficiently quickly to ensure uninterrupted production of a sound weld).

A human being is a complex sensory, interpretation, computational and implementation system! Regrettably this complexity also permits considerable variability in the manner in which work is performed (no two people work in precisely the same way) and in the results achieved. It is unlikely that a welder can totally reproduce a given weld. Such inconsistencies are reflected in job time and quality which may, in turn, give rise to production scheduling and even industrial relations problems. Additionally, the productivity of human labour is limited and its cost, which is inflation linked, is attributed to product prime cost. Output and cost have been incentives to develop the improved control systems necessary for implementing automated manufacturing systems.

Individual machines

The search for greater output and more predictability (consistency) through system control has led to development of automatic machines (systems) which attempt to replicate the actions of a human welder continuously without error. The automated TIG welding system developed at TWI as part of the ESPRIT 9 Project ^[3] ( Fig.2) has three sensory systems, the most important being the vision sensory system, and has been given fundamental 'control rules' which enable it to replicate, in part, human actions.

The model of this system has, as its starting point, the same weld quality and procedure rules as its human counterpart, but it has also been made to cover many other functions, and their associated data, which exist in mechanised systems, e.g. if the joint is not in the field of view of the vision system, search for the joint (function) using visual information (data). The complete model, which is complex, needed to be expressed succinctly and unambiguously so that it could be checked for accuracy by specialists.

This expression took the form of drawings diagrams; Fig.3 shows how IDEF _o methodology was used to express the model of the system in question and how each function was decomposed to levels which ensured that all necessary aspects were addressed. A more detailed explanation of systems analysis techniques is given in an earlier Bulletin article. ^[4]

The functions (boxes) of this model define 'what is required to happen' to ensure that the total system performs as planned, whilst the lines on the model define data flowing between these functions. To allow mechanised control, many of the 'functions' had to be converted into computer understandable algorithms.

For example, the function 'control arc current' (I) becomes:
I = C1 + C2 x prep width (C1 and C2 being constants defined by experiment). (This algorithm is representative of the type derived by the parameter fixing technique developed in the ESPRIT 9 Project. ^[4] .)

When operated, the system automatically goes through the routines of

1. Observing (via its various sensory systems) what is happening;
2. Comparing observations (data fed back from the sensors) with the functions within the model of the system, i.e. does I = C1 + C2 x prep width? and, based on rules which have been embedded in the algorithms, deciding whether anydifferences between observations and model necessitate changes;
3. Determining adjustments (type and extent) to be made in accordance with other inbuilt algorithms;
4. Implementing necessary adjustments through its various servo-activation circuits and mechanisms.

This system, which has limited powers of interpretation, replicates a human welder and can react in real time, but is complex and requires considerable computing power.

Manufacturing cells

Given that automation of individual machines/processes has been achieved, it is reasonable to expect that several machines (or functional systems) may be coupled to form a 'cell', which will manufacture major parts of a product. This has happened in 'flow line' manufacturing, where product and process are fixed, thus allowing the manufacturing operations and their sequence to be fixed and controlled. In batch manufacturing operations, where fixed product/process are unlikely, group technology (GT) cells, which produce a limited range of similar components, are used. More recently, flexible manufacturing systems (FMSs), which are capable of operating on a much wider range of components, have become available.

These are 'integrated' cells, in so far that products and data (concerning product and process) are passed around the various elements of the cell automatically. Each machine is controlled as described earlier; however, their interaction, the critical aspect of any cell, is controlled by a cell controller. In simple terms the cell controller may be considered as the shop foreman; it directs material flow, labour and machine use so as to ensure optimum efficiency. In the case of small flow lines and some simple GT cells a programmable logic controller (PLC), essentially a small computer, will suffice as a cell control; however, large flow lines, most GT cells and all FMSs require a larger computer for this function.

The welding cell ( Fig.4) developed at TWI during ESPRIT 595 ^[5] is just such an FMS. The model of the cell (developed by TWI staff and expressed in the DAFNE ( Copyright Italsiel SpA, Rome) methodology) records precisely what each of its elements (CAD, robot programming, robot control) must do. Control of individual elements is accomplished by the commercial packages purchased with them (e.g. accuracy of dimensional data from the CAD system, positional control of the robot), but the cell controller, which is not a proprietary package, needed to be modelled and built by TWI, according to its needs. 

It is possible to build models describing the operational strategies which cell controllers must use for specific ranges of duty. Whilst a cell operates within its range all is well, but when the unforeseen occurs (e.g. out of tolerance piece parts) such models are unlikely to be able to represent human thinking in respect of implementing alternative manufacturing methods; at best they can cater only for 'safe' systems shut down.

There are numerous possibilities of unforeseen events and alternative manufacturing methods which may occur in a factory and to which humans are able, through their experience, to provide a sensible reaction.

Even with present day computing power, it is impracticable to generate the software necessary to allow computerised cell controllers to match humans in this sphere.

Within the limits to which it is possible to model the 'cell controller, it is usually possible to control the cell.

Experience in developing TWI's weld cell indicates that modelling and building (i.e. generating software code) a realistic cell controller is a substantial task. Even with a restricted product range and making extensive use of external systems to supply data which, in a jobbing shop, the foreman might reasonably be expected to provide (e.g. scheduling and process planning), it has not been possible to provide more than rudimentary capability.

Areas and enterprises

Manufacturing enterprises are generally organised as a management hierarchy, ^[6,7] the supervisor at each level (foreman, manager, director) acting as a cell controller ( Fig.5).

The higher the level in the organisation, the greater the extent to which the cell controller must provide intuitive rather than mechanistic decisions, and the more difficult becomes the task of modelling the interactive functions of the cell.

As the number of interactive functions which a cell controller must oversee increases, so does the computing power required to handle the necessary operational packages and control rules. A reasonable PC (HP 310) satisfies the needs of TWI's welding cell; however, area plant and enterprise control demand escalating levels of computing power.

Today's control technology (i.e. systems analysis and modelling techniques, sensory systems, data transmission and computing power) has proved adequate for automated operation of a wide range of machines and manufacturing cells. ^[8,9] In specific instances this technology has been successfully extended to facilitate automation of complete operational areas and plants (e.g. chemical processing, motor vehicle manufacture): however, the complexity and value of such plants usually demand a human presence to meet unforeseen (thus unmodelled and uncontrolled) circumstances. Total automation of a manufacturing enterprise must remain incomplete until it can be modelled, and systems to simulate human thinking, such as expert systems, have been perfected.

The future

• Improved product consistency and quality;
• Reduced manufacturing costs.

These are key elements for survival in a competitive market place, and are likely to be the motivating force for achievement of increasing levels of control.

Improvements in computing technology (e.g. the transputer, the application of expert systems) and in our ability to generate sound models of the required manufacturing systems will reduce some of the financial and technical barriers to development of the controls necessary for integrated manufacturing systems.

In the short to medium term control is likely to limit the development of manufacturing systems and a human will, therefore, remain the ultimate cell controller. To aid him/her in this role there will be an increasing array of computer assistance (e.g. manufacturing resource planning (MRP), computer aided process planning (CAPP), shop floor data collection (SFDC)) which will provide the data required to make the decisions which 'control' the system.

With the foregoing in mind, The Manufacturing Consultancy at TWI is exploring a number of the elements necessary to advance control technology, e.g. tools for developing systems models, expert systems/neural networks, some of the popular computer aided planning tools and manufacturing networking protocols.

References