| I'm in univerity research on SMT assembly systems. Especialy Simulation and Optimisation using OR-Methods. We developed flexible simulation models of various assembly-machines and use them for performance estimation (Components per hour) under different environements. (Planning methods, PCB-structure, Machine-konfiguration). | My question: | How do you (the people out there in the rel world) plan and estimate the performance of your equipment. (placementrate) | - when planning complete new lines | - after changing one machine | - after changing the pcb | - after changing planning methods | - when planning the needed capacities for a new order. | | Do you use and trust maufacturers software (Philips, Siemens ..)or third party Software (Exaline, Unicam...) ? | Do you just reduce the promised sales-man rate by e.g. 30% ? | What do you do ? | How good is the resulting data ? | Where are the big questionmarks ? | | Long question, but any hint to parts of it is appreciated. | Thanks | Andreas | Andreas,
Performance measurements and estimation of any SMT is one of those area's that is slightly open to interpretation depending on who you are and if your a manager or not plus whether or not the figures make you look good or bad. You know what they say about statistics, lies dam lies!. Ok so there are a number of thing's you have to decide, firstly whhat are you wanting to measure, you need to know this before you can estimate. Some folk's work on Placement's per hour, some use Utilisation some use performace against quote, some folk's use a schedule Vs Cycle time kind of thing basically it's the same thing for different folk's. Management in general ( although I hate to admit it not all) don't want anything that you cant put on a pretty graph that is painstakingly obvious to a 5 year old if it's good or bad, they don't want the reason's cos it's the job of the engineer to sort the problems. They want to know how much money they're �,$ or mark is earning them per hour on a machine be it Fuji, Siemens or Phillips. For planning purposes you need a measure of through put, that can either be placement's per hour ( if your smart you can use the theoretical placement rate of the machine to automatically calculate the utilisation of the machine as well) or pure boards per hour. You schedule against an allowance of say 10%for thing's like feed changes, nozzle problems that kind of stuff then monitor hourly how the line runs and flag when it misses targets. For initially estimating how long a product will take to run you've got standard minute values for how long each type of machine takes to put down a type of component so it's a matter of counting component's and then multipying it up, of course the reality if slightly longer cos you have got feeder positions and table movement's to think about and it just start's to mess with your head. Unless your a massive OEM or a CEM that's landed the contract of your life it's not likely you would exclusivle run 1 board down a line 100% of the time so all the scheduling is done by how long it takes to build a particualr board hence you know how long your line is going to be tied up to build X number of boards.
When it come to optimising the lines there are those who look at only a machine and those who look at the bigger picture. Yes you can optimise 1 machine at a time and it'll run quicker, but is the line flowing as good as it could ? answer is probably not so you must optimise the product to the line which mean's all machines and use a Fuji IP3 to place an 0805 if that's what gives you the best performance. In term's of optimisation software Fuji have never been good at it in my opinion. There are a few we have been reviewing, both the ones you mentioned, Exaline, Unicam / Fujicam and Fabmaster some work some don't Fuji's well...it's now been subed out to someone else, need I say more ? Do you believe what the salesman tell you ? well if your a fool then yes, but most of us know better and utter the salesman's dreaded word's "prove it". Fron the one's we tried some gave 30% increase, some 5 some were worse than the std Fuji MCS! A really good system has to look at thing's like loading, machine set up / change over's for product families and common set up's since saving 30 or 40 mins in a set up is a massive amount of time if your in low / medium volume and hi mix stuff. You should be looking to integrate a bit of queuing theory and highlighting if buffering would improve it.
In term's of trials then I say yes you need to have them, preferably you should have the confidence that the optimised program wont bring you line to a grinding halt though, and believe me I've seen that done. It's important to trial a few I mean we were impressed by the 1st one we tried as it gave a 7% increase in productivity, but then we tried another and hit 30%. Now you wont get this on eveyboard so a range of boards have to be tried so you've got to work in some Design for experiment into the program somewhere to prove that it's not a one off thing.
There was some chatter about Exaline / Fujicam and Unicam a while back so maybe chacking the archives would be an idea too.
Don't know if it helps but the proof of the pudding really is trying it for real on someone's SMT lines.
All the best
JohnW
reply »