By Michael Ford, Aegis Software
Industry 4.0 describes both a business model and technology process for factories, to both require and enable them to become super-efficient whilst also being super-flexible.
Traditional approaches to automation in machinery and robotics continue to be successful, but, are mainly focused on specific production operations, leaving critical operational level decision-making on the “to do” list.
Well, if we don’t get to it soon, someone else will do it for us, and it may not be a human.
Will Industry 4.0 make further investment In automation viable?
Looking holistically at the manufacturing business process, we certainly seem to spend more effort moving products around the globe than actually manufacturing them.
This is a real waste of resources, both in the immediate financial sense, and also from an environmental perspective.
Creating factories that supply their products on-demand to a relatively local market, means money that was spent in finished goods logistics and stock-holding, taxes and duties, as well as the potential depreciation in the product value could instead be invested in automation as well as the local human workforce.
As automation is introduced, production costs per process over time decrease as tasks requiring relatively expensive human resources are replaced.
The achievement of a compelling business plan in local manufacturing means that capacity of the factory can expand to meet increased demand.
As expansion happens, more human resources are required, ideally creating an overall positive balance for the humans involved, as not only do they retain their jobs, they can also enjoy wider and more varied responsibilities.
It sounds as though the future of manufacturing is sorted.
Unfortunately, there is a critical flaw to be addressed, one which has remained elusive since the earliest adoption of complex automation.
A good illustration of the issue can be found by looking at Surface Mount Technology (SMT) lines, placing components on to a printed circuit board.
Though performing what might seem to be a simple job, the complexity of programming and setting up materials on these machines has created some legendary challenges in terms of maintaining productivity and equipment utilization, especially as the mix of products has increased.
As conceptually similar automation is now increasingly being considered and applied to the vast majority of assembly manufacturing, there are some lessons from which we can all benefit.
The first impression when looking across an SMT shop-floor tells you everything you need to know about the business of the factory.
Each SMT machine has a light tower, with the light on top usually being the red one, which indicates that the SMT machine has stopped.
Assuming the lights have not been disabled, just counting up the number of red lights across the floor and comparing it to the total number of machines, gives a very reliable indication about the degree of product mix that is being built.
In high-mix scenarios, SMT machines may typically be in their setup phase for up to 80 per cent of the overall available time, leaving only 20 per cent of the time actually making product.
Though this may seem a bit exaggerated, in reality, in some cases the true added value is just half that.
Change is clearly the enemy of production automation.
Though there have been some great technologies developed on these machines over many years to reduce the time needed for the tear-down and set-up between products, there are always hidden compromises.
Rather than setup a unique set of materials for each successive product, materials common to each product could be set in common locations which would reduce or even eliminate material setup times.
Unfortunately, the location of materials is a significant contributor to machine program optimization, severely impacting the throughput performance, as it has to travel further to pick-up common components.
Another idea is to use removable carts of materials that can be slotted in and out of the machine, with setup changes being done in a separate staging area.
Unfortunately, the cost of this in terms of trolleys, feeders, and the very significant impact with multiple duplicate material requirements means that it can be more expensive than the down-time it is trying to prevent.
These are just two of the more simple trade-off scenarios relating to material setup time and throughput efficiency.
When considering machines with multiple heads, stages or even lanes, the whole situation becomes extremely complex.
The engineering intensity that goes into the machine software to resolve these issues has risen exponentially in the last few years, with bespoke complex algorithms from machine vendors attempting to manage whatever product combinations are thrown at them.
This is indicative of the day to day challenges when trying to optimize automation for use in highly flexible environments.
It gets worse.
SMT production lines are made up of many different processes. Few vendors have any real visibility of how the next machines or the prior machines in the line will perform.
The line will run at the speed of the slowest process, bottleneck processes will appear, and change, over time.
This effect can be seen more clearly when lines are run continuously with a single product, made in high volumes.
Each machine in the line is likely to be running between 50 per cent and 95 per cent of its capacity, with only the bottleneck machine being close to 100 per cent.
Another issue that is more clearly seen when running high volumes, is to see how often the line is actually shut-down.
The rate of completion on any continuous production line will never exactly meet the rate needed for delivery to the customer or next process.
Over-producing creates excess intermediate or finished goods stock, which is expensive to own and manage.
Stopping the line periodically may actually be the cheaper option, even if it is not long enough to change the line over to make another product.
The consequences of these decisions are often masked from management scrutiny, as lines are measured based on performance versus schedule.
If the schedule says the machine need not be used, or that changeover time will have to be allocated, this lost opportunity time is likely not included in most production reports or statistics.
Understanding all of these factors when managing lines in very high-mix scenarios explains why reported productivity figures are far higher than the real-world machine utilization, which crucially means that actual performance and hence return on investment will not be achieved within the time expected, if in fact, at all.
All of these issues together, represent quite a complex planning environment, which is going to become typical across all areas of assembly as further automation is introduced.
In terms of help from software systems, in the SMT planning world, it is a shock for most people to realize that in the vast majority of SMT factories such planning is done in Excel.
There is good reason for this, as traditional planning tools, even the most complex of “finite” planners, have two fundamental flaws.
Firstly, they do not have awareness of the real-time situation on the shop-floor, so basically they are blind to the current and historical status and performance, except for periodic, usually manual, data entry.
Secondly, these generic planning tools don’t have enough specific knowledge of individual processes to be able to reliably know the expected performance of machines given the specific products and mix of products to produce, including program optimization effectiveness, material tear-down and setup time and any other transitional times, such as programming, QA check, conveyor adjustment, PCB support positioning, oven heating or cooling times.
Without this critical information, the expected minimum processing time and average processing time per production unit cannot accurately be calculated.
Existing traditional planning systems are therefore effectively both blind and have their hands tied behind their backs.
Time for a new approach: The digital factory
Intelligence, whether biological or artificial, can only work based on available information.
With the emergence of IoT technologies, such as IPC’s Connected Factory Exchange (CFX) IoT standard, accurate, detailed and timely information flows freely across and between all machines and other processes in the factory, including of course being available to modern digitally orientated MES software that may wish to make use of the information for planning.
Taking the example of what is included in the CFX specification, visibility of current and historical performance can be understood perfectly.
Engineering information including the complete digital product – process model, as well as all needed tools and resources, can be exchanged.
This unprecedented capability means that the software’s “eyes” can be opened and “hands” untied.
The use of software in the digital factory is mandatory, as there is far too much continuous information for a human to take in.
Software orientated for the digital factory takes the continuous data and converts it in to actionable information.
We already see the most modern digital MES solutions utilize data streams from the shop-floor, creating value to make the performance and status of the operation fully visible, including indications of constraints, potential effects and consequences when considering any kind of change, for example when a new order comes into the factory and a practical delivery commitment needs to be made.
As more and more processes adopt CFX across assembly, the more effective and dependable these software algorithms will become.
Steps towards the birth of manufacturing AI
There are two transitions that factory management software will go through as the quality and availability of digital data progresses.
The first is to reach the critical mass of data such that decisions can be made that no longer need human involvement.
Until that point is reached, information is utilized by software to provide a kind of digital visibility of performance and opportunity, such that a responsible person can take informed decisions very quickly, thereby implementing actions a great deal sooner.
The software considers a great many more options and constraints in a much shorter time than by non-digital methods, meaning that changes can be made often and will be more subtle with less “undoing” of the previous plan, ensuring that the operation continues to work in an optimum condition, with a smoother operational flow.
In essence, it becomes possible to perform value-stream mapping for the factory as a whole, in real-time.
An important aspect at this stage is that in addition to the regular reporting of information such as machine status, progress and issues, the machine software should also actively interact with the factory level software, to provide expected cycle times obtained through the simulation and optimization of individually assigned products, as well as product combinations that the factory level software would like to consider.
In addition, opportunities are created for the machine software to be enhanced to utilize information available from the factory level and other processes, to create more advanced optimization algorithms for programming and load balancing.
The second transition comes with the introduction of true artificial intelligence.
As part of getting to the stage where human involvement is no longer required for most decisions to be made, software algorithms will have been created that become quite refined and dependable.
Each algorithm works by trying many different potential solutions for the given problem which are each measured in terms of success through some kind of scoring methodology.
The key difference with the AI, is that the scoring mechanism can be self-adjusted by the AI, as it compares real-world effects and trends with expectations, then seeks to find and use additional information that are influencers on the unexpected effects and trends that the original algorithms did not consider.
In the real world of manufacturing, it is very important that the AI is provided with the benefit of recorded “experience”, accumulated through the development of the initial algorithms, at the time it starts.
The AI can then kick-off at least as good as the algorithms performed prior to AI modification.
In terms of the experience on the shop-floor, the transition to an AI itself is therefore not physically seen as such a big step.
AI digital end-game
As in popular science fiction movies about AI, there is little chance that the introduction of an AI will be the end of the story.
Having achieved a digitalized factory operation run by an AI, there will be continued human desire to expand the scope and reach of the AI to further enhance factory operation.
This can include application to the supply-chain, where materials orders, deliveries and logistics become managed by the AI, as well as quality management, where assessment of the unique circumstances that lead to any defect can be quickly and precisely considered, so as to create an immediately effective, active quality control.
The AI may also be used to manage resources, including people. Rather than having production operators simply assigned to a single repetitive task, the AI can manage their time quite dynamically so as to provide a far more flexible support for the factory operation.
This may include, for example, some complex assembly, followed by quality control, then helping with a material issue, or even performing machine maintenance.
Though the AI is capable of providing the scheduling and technical data associated with each task, most likely delivered through augmented reality technology, this brings a higher degree of skill, diversity and creativity for production engineers and operators.
In fact, this all makes the AI driven digital factory of the future, quite an interesting place to be.
About the author: Michael Ford is the European marketing director at Aegis Software Corporation. He has more than 30 years of experience working in the electronics assembly manufacturing industries.
