Flexible Synchronization of a Stepping-motor using FPGA

Transcription

1 FUSE AE Page 1 FUSE DEMONSTRATOR DOCUMENT Flexible Synchronization of a Stepping-motor using FPGA

2 FUSE AE Page 2 Abstract: Gilbos is a company with 220 people. Its core business is the design, manufacture and sales of textile machinery, more specifically winders. All the design for the machines is done in house: including concepts, mechanics, and electronics. The objective of the AE was to lower the manufacturing cost of a precision winder by replacing a servo-controller, motor, and precision encoder by a stepping motor controlled by dedicated electronics. The technology, chosen for the controller was an FPGA. The total cost for the development was 40 K ECU. The project started on the 15 th February '97, and was concluded on the 30 th September, '98 taking 7.5 months, as opposed to the 6 months planned. The cost of the new winders was reduced considerably. The component cost of the precision-winding option that amounted to 15% of the winding unit, will be reduced by about 34%. This will enforce the market position of the company. The payback period will be less than one year, A return-of-investment of 350% will be realised during the first 3 years (the expected life-time of the product). The project was a success to the company and first production is planned for Q3, Manufacture, assembly and test will be done within the company. The most important lessons learned by Gilbos, which resulted in the 1.5 month delay were: The FPGA-size needed was larger than originally expected, leading to long lead-times to order components to allow implementation of the prototypes. In future more time should be allowed at specification phase. The high clock-frequency of the FPGA (40MHz) caused timing problems in the test set-up that had to be fixed during testing. This should have been considered as a potential problem and some contingency planned in to the project. Measurements in the field required some changes in the specification, and had not been planned for. Keywords: FPGA, stepping-motor, winders, textile, synchronization 1. Company name and address GILBOS NV Grote Baan 10 B-9310 HERDERSEM - AALST Tel 053/ Fax 053/

3 FUSE AE Page 3 2. Company size Gilbos employs approximately 220 people. Three people in the R&D department have experience with microelectronics, more specific with the design of electronic control systems for textile machinery. This design includes: - design of the concept; - selection of the sensors, drives and PLC's; - design of PCB's to extend capabilities of the PLC control: D/A and A/D conversion, pulse dividers, amplifiers, PNP/NPN converters, electronic interfaces with galvanic separation by means of opto-couplers and relays, user interfaces with 7 segment displays,.. A schematic capturing software (UltiCap) and a PCB development software (UltiBoard) is used to design the board. The necessary equipment to produce prototype PCB's in the electronics lab is available. - development of the PLC software to control the machine, including the implementation of several protocols for serial communication (RS232 - RS485); - development of software for industrial Men-Machine Interfaces (MMI); - development of Visual Basic programs for Personal Computers used as MMI; The following Gilbos employees were involved in the AE: - Kurt Muylaert: R&D Engineer - Christian Van Hautte: R&D Manager For the design and production of some mechanical parts that were needed to test the prototype, several other employees contributed to the AE. 3. Company business description Gilbos' core business is the design, manufacture and sale of textile machinery, more specifically winders. Design: All the design for the machines is done in house: concepts, mechanics, electronics,... The necessary tools like CAD are available to accomplish these tasks successfully. Manufacture: A large amount of the parts for the machines that are not commercially available on the market, are produced in house. One division is specialised in sheetmetal parts and uses up to date machinery like lasercutters, punchers, bending machines and a high quality paint shop. A second division is specialised in milling and turning operations. A subcontractor produces most of the casted and composite parts but the machining and finishing is carried out by Gilbos. Sale: Gilbos has agents who represent the company in the most important textile countries throughout the world. A sister company called Gilbos of America does the representation in the USA. The marketing and sales department provides the agents with pricing and technical information and sales support. Gilbos' customers are mainly major carpet manufacturers such as Beaulieu (B), Balta(B), Shaw (US), Saudi Carpet, Milliken (US, UK), Victoria Carpet (Australia). Installation and maintenance: Gilbos technicians travel around the world to install and maintain new and existing machinery. A service and spare parts department provides the

4 FUSE AE Page 4 customers with support. For the US market, installation and maintenance is carried out by Gilbos of America. The textile business knows a lot of economical ups and downs and forced the company to diversify. The know-how and available production facilities are also used for other activities, where Gilbos acts as a subcontractor outside the textile machinery industry. About 25% of the business is created by producing mechanical parts and assembling apparatus for other companies. 4. Company markets and competitive position at the start of AE Gilbos' core business is the design, manufacture and sale of textile machinery. Gilbos operates on a niche market: Winders for coarse yarns (yarns for carpets and for technical textiles). In some market niches, Gilbos is the market leader. However, there is growing competition from competitors with production facilities in low wage countries. During the last 15 years, Gilbos has established a market leadership by being the only manufacturer, worldwide, able to offer winders for each application within this niche market. This was achieved by creating a whole range of different types of machines. One of these machines is a precision winder, called PLS. Picture 1: 3 modules with each 2 precision winder units. Packages made on such a winder typically have a higher density and a better unwinding behaviour than conventional packages. Unfortunately, production of precision wound

5 FUSE AE Page 5 packages requires an extra operation on these special winders, making this process rather expensive. For the last 3 years, Gilbos has worked on a development program for a new family of coarse yarn winders. The aim of the development was: to create several new applications for coarse yarn winders, in order to extend its market; to improve the performance of the machines in order to extend its market-share; to make a modular design, allowing to manufacture more efficiently the whole variety of different types of machines. In this new family, the drive system of the machines also needed to be a modular design, in order to offer precision winding as an option on all the variants, i.e. without an extra operation on a special machine. In the current design, the machines are of a modular design. The precision winding drive module contains an asynchronous servomotor, which performs excellently, but which is rather expensive. No further improvements could be made without the introduction of new micro-electronics technology. By using a FPGA combined with a stepping motor, costs can be reduced considerably, making the precision winding option attractive for the clients. Not only will this lead to relevant cost savings; the cost for the precision winding can be decreased by 35%but by being able to offer precision winding at a very attractive price on all variants, Gilbos' position in the market will clearly be reinforced and even higher market shares will be obtained. 5. Product to be improved and its industrial sectors Gilbos winders are used in the textile industry, mainly to wind coarse yarns (yarns for carpets and for technical textiles). A winder winds yarn onto bobbins; the yarn is hereby wound around the rotating bobbin, while a traverse guide moves the yarn up and down along the bobbins surface. The resulting bobbins are used in tufting or weaving creels, or just to transport the yarn from one production process step or location to another. Most of the winders sold by Gilbos up to now, are so called random winders, whereby the speed of the traverse guide is not very critical. One of Gilbos winders however is a precision winder. A precision winder produces superior bobbins, characterised by: Better unwinding properties, resulting in higher efficiencies during the subsequent processes; Higher densities, resulting in less handling and transportation costs; Better yarn quality.

6 FUSE AE Page 6 On a precision winder, the speed (rpm) of the traverse guide has to bear a constant but very accurate ratio to the rpm of the bobbin, and this ratio has to be adjusted in function of the yarn to be wound. This ratio is called the winding ratio. In the past, Gilbos realised this by means of a mechanical transmission with T-belts between the bobbin holder and the yarn traverse guide. Of the existing product, Gilbos has actually two variants, called PLS10 and PLS10D respectively. The PLS10/PLS10D family was launched on the market during the International Textile Machinery Exhibition ITMA in Gilbos owns a European Patent (Nr ) and a USA Patent (Nr. 4,763,849) on the technology used for these winders. For a new design, Gilbos has preferred to use an electronic solution instead of the mechanical transmission, whereby a first motor drives the bobbin (=master), and a second motor drives the traverse guide (=slave), and whereby the slave follows very accurately the master according to a remote controllable ratio. Picture 2: Detail of servo-controlled precision winder unit. This electronic solution makes a precision winder more flexible, allowing using precision winding efficiently, even in frequently varying circumstances. Several prototypes of this new design have been made, and successful field-tests have been organised. The machines performed well, and the clients were extremely pleased with the concept.

7 FUSE AE Page 7 On these prototypes, a precision encoder detected the rpm of the bobbin, while an asynchronous servomotor drove the traverse guide. Remote setting of the winding ratio was done via a RS485 interface between the machine control system and the servocontroller. The servo-controller is a piece of standard equipment (Control Techniques - UK), available on the market. The cost for the encoder and the servomotor and the servocontroller is 1475 ECU. Whilst the cost is rather high, the competition does not have precision winders in their competing products. Therefore, if a customer requires this option then they have to purchase it from Gilbos. However at such a high cost, many customers simply do without. The ratio between the speed (rpm) of the master axis (bobbin) and the speed of the slave axis (yarn guide) has to be accurate to within five millionths. During the build-up of the package, the master axis gradually slows down. Depending on the kind of yarn and the winding speed, the ratio between the master and the slave axis is changed every 2 to 5 minutes. Encoder on master axis 1000 ppr PLC Ratio (RS485) Servo CTRL Servo motor on slave axis Resolver feedback Block-diagram of control for precision winder with servo-motor. Technically, (performance, size, power), this solution works very well, but unfortunately it's relatively high cost has limited its commercial success. Cost reduction was the only driver to improve the product. 6. Description of the technical improvements Gilbos wanted to overcome the disadvantage of the relatively high cost of the servo controller and motor and the precision encoder, but without losing the important advantage of flexibility. Therefore, an alternative electronic solution was worked out. This was based on a speed controller with a FPGA, which transforms a pulstrain from a low-cost sensor on the bobbin axis, into a high precision pulstrain controlling the speed of a stepping motor. This stepping motor is driving the traverse guide. Furthermore, this FPGA is able to communicate with the machine control system in order to allow remote setting of the winding ratio. Sensor on master axis 10 ppr PLC Ratio (RS485) Pulstrain Stepping motor on slave axis Speed Controller Stepping motor controller Block-diagram of control for precision winder with stepping motor.

8 FUSE AE Page 8 For precise control of the movement of a stepping motor, a high precision pulse train has to be fed to the stepping motor controller. The required precision for this application can, in the worst case, be as high as within Hz. The input data for the speed controller are the pulse train generated by the master axis and the required ratio between the number of revolutions of the master and the slave axis. C SERIAL INTERFACE PULSE TRAIN FROM MASTER AXIS RATIO RS485 INTERFACE PERIOD MEASUREMENT RATIO DIVIDER MASTER AXIS SPEED REQUESTED SLAVE SPEED PWM GENERATOR SLAVE AXIS PULSE TRAIN OUTPUT STAGE Speed Controller TO STEPPING MOTOR CONTROLLER Detailed block-diagram of the speed controller The number of revolutions of the master axis can be as high as 6500 rpm and decreases continuously during the winding process. The speed of this axis has to be measured continuously and very precisely. The ratio between the number of revolutions of the master and the slave axis is sent by a PLC trough an RS485 serial interface. This ratio can be any multiple of between and (e.g ). The pulstrain that is generated by the speed controller is sent directly to the stepping motor controller. The frequency of this pulse train needs to be updated continuously to maintain the precise relation between the master and the slave axis speed and position. The main component on the speed controller is a gate, 208 pin FPGA that takes care of the period measurement, the division and the pulse generation and the serial communication.

9 FUSE AE Page 9 Picture 3: prototype PCB with the speed controller. Picture 4: detail of the stepping motor The main product improvement is the reduction of the cost of the electronic parts needed to achieve the flexibility of a servo system: - The speed controller makes the high precision encoder (1000 pulses per rotation) that monitors the number of revolutions of the package, superfluous. It can be replace by a

10 FUSE AE Page 10 simpler sensor (10 ppr) that is a cheaper and less sensitive piece of equipment. - A stepping motor and controller, combined with the FPGA base speed controller is cheaper then a comparable servo system. ECU I. Existing drive system : Asynchronous servo motor + servo drive : 1375 High resolution encoder : 100 Total component cost : 1475 II. New technology : FPGA control : 200 Stepping motor (incl. control and power supply) : 700 Sensor master axis : 75 Total component cost : 975 III. Component savings per winding unit : 500 As an average, a complete winding unit is sold on the market for around 10 K ECU. It is clear that therefore: - the cost of purchase of the key-components of the servodrive system is relevant (cost of purchase of key components = 15 % of the sale price of an average winding unit) ; - the cost of the option precision winding will reduce with approximately 34 % by introducing the FPGA technology. In this way, the new concept of the Gilbos precision winder is not only technically superior in comparison with the existing (mechanical) design, but also financially justifiable. 7. Choices and rationale for the selected technologies, tools and methodologies A slave axis has to be synchronised very precisely with a master axis. The master axis provides a pulse train, the slave axis is controlled by a second pulse train that is related to the master pulse train. The relation can be any multiple of between and The easiest and most flexible way to realise the divider is by means of software. Software would allow to take the decimal part of the result after the division into account during generating the pulse train. Gilbos have written a PLC program that, using high-speed cards is generating a pulse train that meets the levels of precision requested. Due to the scan time needed by the PLC to execute its program, this solution only works for low speeds of the master axis and can therefore not be used in an industrial application. With the PLC, a yarn speed of 50 meter per sec could be achieved, whereas a minimum speed of 1200 meter per second is required. - The next step would be to implement the same program on a micro-controller. Gilbos have only very little experience with micro-controllers, but calculations are showing that it is not possible to achieve the requested frequencies at the requested precision by using a micro-controller. Since a micro-controller needs to do all its calculations in a serial way, the required throughput of 40 MHz to control the stepping motor cannot be

11 FUSE AE Page 11 achieved. The performance problem can be compared with the performance problem of the PLC. - The previous considerations brought Gilbos to the conclusion that this problem should be resolved with a hardware solution. Implementing a divider with commercial available IC's seems very unlikely due to the requested precision and the intended (small) size of the board. Developing an ASIC for this application was not a valid option because of the limited quantities ( per year) and the low flexibility. Gilbos needed flexibility to be able to customise the control electronics for each customer if necessary and they wanted to maintain this flexibility in every new design. - The lead-times for ASICs developments are quite long 12 week prototype manufacturing cycle and the NRE cost for ASICs is much higher than for FPGA. Furthermore, only very small production volumes are required maximum a few 100 units per year. FPGA is the right solution for this project. Compared to a micro-controller, all processing can be done in parallel, in special dedicated hardware. The computing throughput in the FPGA is much higher than can be achieved with the serial processing in a micro-controller. The FPGA has a high flexibility compared to an ASIC, it is easy to make small changes in the netlist, and the cost for implementing a new prototype is low, since it can be done inhouse, without having the lead-time for mask-manufacturing and Silicon prototyping in the Foundry. Since the annual volumes are low, the unit cost for the FPGA's (153 ECU) is not very important, compared to the economic saving realised by replacing the expensive servo-controller, motor and precision encoder, by lower cost encoders and steppingmotors. The total saving of 35% can be realised. After analysing the problem, the TTN confirmed that the FPGA technology was indeed the correct technology to be used. - The lead-times for ASICs developments are quite long - 12 week prototype manufacturing cycle - and the NRE cost for ASICs is much higher than for FPGA. Furthermore, only very small production volumes are required - maximum a few 100 units per year. - The FPGA was developed using the VHDL methodology. This high-level hardware description language allowed Gilbos to understand the construction of the design and will enable them to develop new FPGA's without expert help in the future. - A Xilinx FPGA of the family 4000 will be used, because Barco - Silex (the subcontractor) is very familiar with this component. The choice of the Xilinx component leads to the use of Xilinx design environment for the programming of the device. This environment allows design entry in VHDL, simulation, and testing of the programmed FPGA. - The FPGA loads its software from a serial EPROM. The advantage of this configuration is the flexibility it offers to install modifications without replacing the FPGA. - Testing and evaluation was done in 'real life' on a machine together with Silex and Gilbos.

12 FUSE AE Page Expertise and experience in microelectronics of the company and the staff allocated to the project - Gilbos designs the electronic control systems for its textile machinery in house. The staff involved in this FPGA development had experience in: - design of the concept; - selection of the sensors, drives and PLC's; - design of PCB's to extend capabilities of the PLC control: D/A and A/D conversion, pulse dividers, amplifiers, PNP/NPN converters, electronic interfaces with galvanic separation by means of opto-couplers and relays, user interfaces with 7 segment displays,... - a schematic capturing software (UltiCap) and a PCB development software (UltiBoard), which were used to design the board. - producing prototype PCB's in the electronics lab. - development of the PLC software to control the machine, including the implementation of several protocols for serial communication (RS232 - RS485); - development of software for industrial Man-Machine Interfaces (MMI); - development of Visual Basic programs for Personal Computers used as MMI; The technology used in this AE and in which Gilbos is a First User, is (Field Programmable Gate Array) FPGA. None of the technicians at Gilbos has ever worked with this technology before, so a partner was needed that had the required know how for the design of the FPGA. Also Gilbos had no previous experience in the use of VHDL for the design of electronic circuits. 9. Workplan and rationale. Gantt Chart with effort of Gilbos in working days. Months Management Training System analysis Specification Design Prototype & Test Total Gantt Chart: Shaded fields are actual, un-shaded italic fields were planned.

13 FUSE AE Page 13 Tasks Effort Gilbos Cost Subcontractor Task 1 Management 7 days Nil Task 2 Training 7 days Nil Task 3 System Analysis 6 days Nil Task 4 Specification 1 day 4690 ECU Task 5 Design 7 days 9380 ECU Task 6 Prototype & Test 16 days 2340 ECU Total: 44 days ECU There was an additional cost of 1200 ECU for training, organized by the Xilinx component importer, and of 1391 ECU for components, test PCBs and other consumables. The total cost for the development was 37.2 K ECU, including Gilbos' manpower. Since there was a fixed price quotation from the subcontractor, there was no difference between actual and planned cost. The duration of the project exceeded the planned duration by 1.5 months. This delay was caused by long lead-times to get the FPGA-components and by some minor problems during testing (mainly due to high clock frequency - 40 MHz - and to changes required in the behavior of the system during fast acceleration). The role of the TTN consisted mainly in assisting the communication between the subcontractor and the first user. This was done by asking questions to the subcontractor, encouraging him to give technical background information, and then making sure that both partners really understood each other, rather than only believe they had understood. Task 1. Project Management All project management was done by Mr. Van Hautte, the Gilbos R&D Manager. Project management included project set-up, selection of the subcontractor, internal communication towards general management, communication with the subcontractor, follow-up and co-ordination and planning. Task 2. Training Two Gilbos engineers attended a VHDL course at SEI in the Netherlands, the Benelux importer of Xilinx FPGA's. This course was sufficient to learn to understand the design flow used by the subcontractor (VHDL, simulation, programming of the FPGA) and to allow Gilbos to interact with the subcontractor. This has been useful during the test phase of the circuit when a few changes in the algorithm of the behaviour during acceleration had to be made. It will also allow Gilbos to initiate new FPGA developments internally for future projects. Task 3. System Analysis: Evaluation of the technical aspects In this stage, tests were done with stepping motors to find the most suitable settings for the stepping motor controller. These setting have an implication on the output frequency generated by the speed controller. These tests showed that the idea of making a precision wound package with a stepping motor mechanism was valid.

14 FUSE AE Page 14 A meeting with Barco Silex was organised to explain and to show the principles of precision winding, the existing machines and the purpose of the FPGA that needed to be designed. The result of this stage was a document, prepared by Gilbos that contained the technical specifications for the speed controller. In an addendum, the specifications for the RS485 interface were added. The tests also showed that fast speed changes might cause the stepping motor to stop following the applied pulse train. The conclusion was drawn that in the FPGA, a speed change limitation should be implemented. Task 4. Specifications and general block schematic During this stage, Barco Silex examined the technical aspects of the speed controller and created a document that contains a block diagram of the FPGA and an explanation of these different blocks that were needed to accomplish the requested function: speed measurement, a RS485 interface, a decimal to binary conversion, a multiplier, an output frequency generator and an acceleration computing block. The specification document was discussed with Gilbos and used as a basis for the VHDL programming. Task 5. Design of the VHDL code and simulation and evaluation Barco Silex wrote the VHDL code, based on the specifications document that they had previously produced. The designed VHDL code was simulated, using ModelSim. A simulation report was produced by Barco Silex showing the reaction of the FPGA on input parameter changes. In this stage, the acceleration-computing block turned out to be a lot more difficult to design then first expected and caused delay, because this block had to be changed during the testing phase Timing analysis of the design showed timing problems in the blocks that were working on a 40 MHz speed. This high speed is needed to achieve the required precision. These timing problems caused the evaluation phase to be delayed. Task 6. Prototyping and Test or Evaluation To be able to evaluate the design, a test PCB was designed by Gilbos, based on specifications from Barco Silex. A PLC program was written to communicate with the FPGA. There was a delivery problem with the FPGA, which caused delay to the evaluation stage. A few modifications were needed to the PLC software, the PCB (high clock frequency - 40 MHz) and the VHDL code (calculations during acceleration). This was cause of some delay. When everything was build together (PLC, speed controller with FPGA and stepping motor), it took about 2 days to get the system running on an acceptable level. This can be considered as a success, given the fact that everything was new: the PLC software, the PCB and the VHDL code.

15 FUSE AE Page Subcontractor information To find the right subcontractor, the Belgian importers of two major suppliers on the FPGA market (Altera and Xilinx) were contacted and they provided Gilbos with the details of several design houses specialised in FPGA design. After a first consultation of all these design houses, Gilbos requested an offer to realise this project from two different companies. Finally, they selected Barco Silex as a partner for this project because of the simple concept of the proposed solution, the reasonable price of their offer and their excellent reputation in the industry. Barco Silex (Subcontractor) has been active for several years on the high end ASIC market, focused on ASIC, ASSP (Application Specific Standard Products) and FPGA design for industrial parties. More than 120 different design projects have been realised so far, including mixed analogue/digital ASIC's with analogue interfaces with various sensors for the automotive manufacturers, a control with optimised standby power consumption (a few µa) for intelligent sanitary application, integrated circuits for inserting text menus into video images and integrated circuits that are used in the direct control of the engine of locomotives. Usually, VHDL (a hardware description language) is used during the design stage for describing the design and translating it into gates by logic synthesis. If limitations occur in its use (e.g. when speed or chip area are critical), the design is optimised at the gate level part of the synthesiser output. Based on a detailed description of the functions that were expected from the FPGA, Barco Silex made an estimation of the time they would need to carry out the work: specification definition, concept and programmation, and testing and verification. Based on this time estimation, the total cost for the engineering and development was set to a fixed amount of money, 30 % paid by the start of the project, 30 % after validation of the specifications and 40 % after testing and verification. It was agreed that once the project was completed, Gilbos would get the complete source code for the FPGA. Then, if necessary, future changes could be made to the FPGA in house, provided that Gilbos invests in the appropriate tools for designs of this size (28000 gates). Inside Barco Silex, two contact points were established: one commercial contact and one technical contact, so both areas were covered. The information exchange was open minded and very fluent. 11. Barriers perceived by the company in the first issue of the AE technology In Gilbos' business, it is practically impossible to predict what kind of machinery will be sold in the next 12 months. They operate in a niche market, and they have a whole range of different types of machines so they cover each application in this niche market. Because of this unpredictability in sales, it is difficult to take a decision on what to design or improve next. Gilbos was aware of the capabilities of FPGA and was convinced that FPGA could help to reduce the total cost of several electronic controls. In the textile machinery market, it is very hard to predict the success of a new development. Gilbos have always been delaying the first step into this technology because of the cost of the investment combined with the

16 FUSE AE Page 16 technical risk that are inherent with the first project. There are several FPGA suppliers with a large product range on the market, and the choice of the supplier and the type of FPGA depends upon the complexity of the solution that has to be implemented. The more complex a FPGA, the higher the price, so picking a wrong model has a large influence on the price of the final design. Once the decision on the supplier and the correct type is taken, a choice has to be made about the development tool that is appropriate to implement the solution in the FPGA. Development tools are rather expensive and buying the wrong tool could mean a considerable investment loss. Finally, before Gilbos could start implementing a solution with a FPGA, one or more of the R&D engineers need an intensive training in this technology. It is clear that Gibos could not gather the necessary knowledge without the help of an expert in the FPGA technology, and usually expertise is very expensive. During the first project, Gilbos wanted to gather enough knowledge to be capable of implementing solutions for other problems in the FPGA technology without the help of external experts. 12. Steps taken to overcome the barriers and arrive at an improved product The FUSE project offered Gilbos an excellent opportunity to introduce the FPGA technology into the electronic control systems. After analysing the problem, the TTN confirmed that the FPGA technology was indeed the correct technology to be used. Gilbos have found expertise help with Barco Silex as a subcontractor. After analysing the specifications and based on their experience, they were able to advise on the type of FPGA that should be used to implement the final solution. Two engineers at Gilbos attended training in FPGA programming and the tools that are required during the design stage. Although it should be clear that it takes more then attending a course to become a successful FPGA developer, the training offered a very good overview on what kind of applications can be implemented with FPGA's and the choice of development tools like VHDL, schematic capturing, state machines, Knowledge and experience acquired. Gilbos was founded over 75 years ago, and is used to subcontracting to and from third parties. On the point of view of management, no additional knowledge or experience was gained from this project. It was interesting to see how Barco Silex approached the project. As with all developments, an analysis had to be made based on the possibilities that are provided by the FPGA. On the first levels of the analysis, it is more important to find a good 'algorithm' to resolve the problem then to know perfectly how to write VHDL code, but care needed to be taken that this final 'algorithm' did not contain parts that could not be implemented in a FPGA. Two Gilbos engineers have attended an FPGA-training. During this training, they got a good understanding of what a FPGA is and how it works. This convinced Gilbos even

17 FUSE AE Page 17 more of how useful these devices can be in reducing the cost of their electronic control systems. During the training, after each theoretical part, an exercise was made using different design tools like VHDL and schematic capturing. The final result of the exercise was a working design. The exercise explained very clear the different steps that need to be taken before the code can be loaded into a FPGA. With the knowledge gained in this training, Gilbos can now make the right choice about the development tools that are needed to modify the existing development and to create FPGA's for other applications. A Xilinx start-up kit will be acquired, and will be used within the company. When needed, this start-up kit will be replaced by a full licence. The level of experience in the FPGA design flow, has allowed Gilbos to interact with the subcontractor, to resolve some problems with the behaviour of the system during fast accelerations, that were detected during test. Because of its acquaintance in using PLCs and flow diagram developments, and the training that has been followed, Gilbos will be able to use the VHDL environment for future products that require the implementation of a.o. state-machines. The test PCB for the FPGA was designed in house. To allow easy repair of their PCB's in the field, Gilbos never use SMD components and all integrated circuits are mounted on sockets. The shape of the FPGA was a SMD HQFP, and they had never used one of these before. They needed some assistance from their PCB manufacturer for the definition of the size of the connection pads and traces. 14. Lessons learned Gilbos has been very carefully in the selection of the subcontractor, and they collected names of possible subcontractors via the distributors of the FPGA components. They then did a survey by contacting those distributors, focusing upon personal contact and communication around the system requirement, the design flow, and they made sure that the subcontractor understood the system and Gilbos' requirements. Communication and involvement are very important. The subcontractor should feel to team with the customer in solving a common problem. Gilbos finally asked 2 of the subcontractors to prepare a final bid. The selection of a good subcontractor is key to success: it is the professional expertise of Gilbos combined with the professional expertise of the subcontractor that can lead to results that each of the partners alone would not be able to achieve, in reasonable time and cost. In this AE Gilbos believe they chose wisely and this is why the project was such a success. The FPGA that was needed to implement the solution was bigger then expected, with around gates needed. One consequence was that the component was more expensive then expected thus reducing the expected savings. Another consequence is the availability of the component: this type of FPGA is not so widely used and was not available at the time Gilbos were ready for testing. This caused a considerable delay. For future applications, a more conservative estimation about the required number of gates is needed, and once this component is fully defined, it should be ordered from the supplier without any delay. Because of the high clock frequency that is used internally to do accurate measurements (40 MHz), during the development stage, timing problems showed up and caused several

18 FUSE AE Page 18 days of delay before the testing could start. There will always be a difference between testing a solution in a lab environment and in 'real life'. Gilbos experienced a few problems during the ramping up and ramping down of the system that required a design modification and caused some extra delay. 15. Resulting product, its industrialization and internal replication Picture 5: prototype board of the speed-control unit. Gilbos are pleased with the result of the project. They still need some time to do more extensive field tests before they start shipping this solution around the world. This testing phase can not be replaced by design, process or management control. If a problem appears after several machines have been installed in the field (throughout the world), it could cause severe damage to their reputation and furthermore have serious financial consequences. Gilbos wants to conduct the field test in the plant of a Belgian customer to be able to supply sufficient support and to get support from Barco Silex without high travelling expenses. Once the field test is successfully concluded, Gilbos will rework the speed controller PCB in house. The necessary components such as the FPGA, serial prom,... will be purchased directly from the Belgian distributor. Gilbos is the complete owner of the design and no royalties should be paid or further commitments fulfilled when the design is used in their machinery. When the reworked PCB is ready, it will replace the existing servo system. Additional investments for industrialisation are the development of a production-quality printed circuit board, and an extensive field test in a production environment of one of the customers. Cost for this is estimated at 4200 ECU. Production test is estimated to be completed end Q2 1999, production start Q3, and first sales Q

19 FUSE AE Page 19 Targeted volume is 70 units for 1999.There is a lot of potential for internal replication of the FPGA technology: - Several of Gilbos' products include one or more sets of rollers which feed the yarn to some yarn treatment device and finally to the winding unit. - The function of the rollers is to assure a correct yarn tension. This is done by controlling the speed difference between two sets of rollers, sometimes by means of feedback of the yarn tension. Up to now, this was done mechanically. - With the experience of the first application, Gilbos will be capable to develop a FPGA solution for this application as well By using this technology, the Gilbos products will become more flexible at an economically justifiable cost. - The actual control systems of Gilbos' products are PLC-based, including a very high number of I/O s (up to 1500/machine). The use of FPGA s could substantially reduce this amount of I/O s. - The fact is that Gilbos' products consist of a number of identical winding units (e.g. 24/machine), each unit requiring a number of I/O s (up to 60/unit). - A substantial part of the outputs only serve for the execution of fixed cycles; these fixed cycles could be controlled by FPGA s. In all cases, substantial cost reductions can be realised, while at the same time these technical solutions fit with Gilbos conceptual strategy: - maximum use of standard available components; - maximum flexibility by developing software within the organisation. 16. Economic impact and improvement in competitive position By using a FPGA combined with a stepping motor, costs can be reduced considerably; the cost saving per winding unit is 500 ECU. As an average, a winding unit is sold on the market for around ECU; it must be clear that: - the cost of purchase of the key-components of the servodrive system is relevant (cost of purchase of key components = 15 % of the sale price of an average winding unit) : - the cost of the option precision winding will reduce with +/- 34 % by introducing the FPGA technology. Not only will this lead to relevant cost savings; by being able to offer precision winding at a very attractive price on all variants, proposer s position in the market will clearly reinforce and even higher market shares will be obtained. The economic impact of the introduction of the FPGA technology in the precision winding concept of Gilbos could be estimated as follows:

20 FUSE AE Page 20 Based on the above and without taking the introduction of FPGA into account, Gilbos expects the evolution of its sales as follows: Basic Sales Growth Basic Sales Volume (K ECU) % % % % % 7000 As indicated, Gilbos expects a market and market-share increase from the availability, at an attractive price, of precision winding on all its product variants: Added Sales Volume (KECU) Basic Sales Volume (KECU) Added Sales Growth Added Sales Volume (KCU) % % % % 1000 Bearing in mind an average profit margin of 5 % on machinery sales, this yields to the following projected added profit contribution, due to the introduction of the FPGA technology for precision winding at Gilbos: Added profits (kecu) The total cost of the experiment was 39,75K ECU. This yields to a direct payback period of less than one year. An additional cost of 4.2 kecu will be needed, to modify the PCB

21 FUSE AE Page 21 for the production units. The Return on investment will be 350 % only during the first 3 years. 3 years is the expected life-time of the product. Indirectly, the experiment offers an even higher ROI. Once Gilbos will have gathered sufficient knowledge and experience with the use of FPGA, other applications of FPGA will follow within the product development of Gilbos (see Internal replication ). Other possible applications are already identified and will be implemented immediately after the successful introduction of the first application. Indeed, the potential cost savings are considerable and these will clearly reinforce the competitive position of Gilbos on the market, by allowing Gilbos to offer automation and flexibility at an attractive price. 17. Target audience for dissemination throughout Europe In this AE, Gilbos, the FU, developed a FPGA to accurately control the speed of stepping motors for yarn winders, operating at speeds of 1200 meter per second. This requires pulse trains up to 40 MHz, excluding the use of PLCs or micro-controllers. The FPGA and stepper motor directly replaced a servo motor system, giving a considerable cost saving with no loss in performance. The FU had no previous experience in FPGA design or programming using VHDL. They were therefore reliant on a good sub contractor to assist them throughout the AE, from the initial specification right through to the testing of the final prototype. The FU also needed training throughout the project in order that they acquired the necessary skills to repeat a similar project. The FU learnt how to cost, plan and choose and work with a suitable sub contractor, in order to produce an FPGA based product. They did not however learn how to programme an FPGA using VHDL since this is something they intend to sub contract on future projects. The FU was successful in this AE, and this was partly attributed to their good choice of sub contractor, with the necessary technical skills, and their own in-house project management skills. Their choice in sub contractor was made by starting with an address list from a general survey starting, initially comprising FPGA dealers. From this list they chose a number who they had initial discussions with, and from this they asked a small number to quote for the work. Having chosen the sub contractor they prepared a fixed price contract which was signed by both parties. During the AE there were some minor delays which were not foreseen at the outset: Details in the specification needed to be changed after testing because of unexpected side effects in the algorithms. This required minor changes in the VHDL and re-programming of the FPGA The FPGA required was larger than originally estimated, leading to some delay due to component procurement time. High frequency signals on PCBs and in interconnect presented problems and needed special attention. This AE represents a simple, well defined application for an FPGA, where the FU had limited previous experience in PLCs and PCB design. It will be of benefit to other companies currently using PLCs, where a digital hardware solution would improve their product. Possible target audiences, which should be interested in the results for this experiment, are certainly all users and manufacturers of equipment in which the speeds of several axes have to be synchronised accurately. More specifically, users and suppliers of machines for processing thin or filamentary material such as textiles, paper, film, packaging materials, cables and wires etc.

22 FUSE AE Page 22 Prodcom codes: 2924 Other general purpose machinery Machinery for different applications 3330 Industrial process control equipment