Cost reducing a design involves redesigning a working system to cut production costs. While the reasons for this are understandable, it often strikes engineers as a betrayal of the project they just spent weeks, if not months, designing. What's worse, is that to be effective at cost reducing a design requires almost starting over and evaluating where things could be done differently. This iterative redesign burns not only time but money, both of which are in short supply these days. What's needed is an approach that optimizes costs at the start of the design.
If we look at other areas that have dealt with this kind of iterative problem, one that comes to mind is the holistic approach of mechatronics. The entire design, in all four disciplines — mechanical, electrical, control theory and microcontrollers — is examined and modeled, as a whole. The design goals associated with each section are based on the entire system's needs. In this way, each section is fine-tuned to work with all of the others, in order to minimize unnecessary overhead.
If this mind-set is applied to cost reduction, we should be able to redistribute the various functions such that the overall design is not only less expensive, but also reduces the number of prototype iterations.
Let us start with a simple example — a sensor, its signal-conditioning chain and the Analog-to-Digital Converter (ADC). Figure 1, above, shows a typical system with filtering, a gain stage and the ADC for conversion of the final value to a digital number. The normal approach is to limit errors in the final value by using components trimmed for low offset and consistent gain, as well as temperature compensation to remove environmental drift. The result is a straightforward design that produces only minimal errors in the final numbers. The problem with this approach is most, if not all, of the components are quite expensive. Temperature compensation and performance trimming require special testing and calibration at the end of the component manufacturer's assembly line. Further, the circuitry required to perform the adjustment must be built into every device in the chain. The result is that the customer now has to pay for the redundant testing and circuitry required to compensate one sensor.
From our new, holistic view of cost reduction, this redundancy is a prime candidate for removal. To accomplish this, we must simply trim and compensate the final value in the microcontroller instead of each of the components, trading a significant component cost for a hopefully smaller final test-and-calibration step for the full system. We will have to add a few functions — such as a thermistor and non-volatile storage — but the temperature sensor and storage will be needed for other functions anyway, so the cost increase should be negligible. Because the microcontroller is designed to handle addition, subtraction and multiplication, additional requirements for compensation and trimming should only slightly increase the software overhead.
Calibration is a dirty word to most production lines, but it can be a reasonable trade-off when compared to the price difference between an uncompensated sensor/signal chain and a fully trimmed/compensated set of components. Further, the microcontroller can also help the process by running the calibration function on multiple channels simultaneously, essentially calibrating the entire system automatically during final test (with temperature cycling) at the end of the production line. It would also be advantageous to have the microcontroller periodically check the calibration of the system, by looking for discrepancies in the sensor outputs while they are in a resting state.
Additionally, it may be possible to calibrate and compensate the system without temperature testing at the end of the production line. If the sensor is needed only to monitor an incremental shift in a given parameter, then offset is not needed. This also eliminates the need to compensate for temperature drift. However, if temperature is a factor in the gain of the sensor/signal chain, then using the components' temperature coefficients and a thermistor input, the microcontroller could perform a “best-guess” compensation, eliminating the requirement for testing completely.
What about driver circuitry — how can a mechatronics' approach save money driving motors, values or solenoids? Regardless of what is being driven, the drivers must be as efficient as possible while providing some range of control. This typically means some form of Pulse-Width Modulation (PWM) drive is needed.
The typical solution is to combine a microcontroller's PWM peripheral with an integrated driver module to create the necessary drive. The integrated drivers are convenient because they contain all of the drive circuitry. Additionally, the actual output MOSFET transistors and the PWM peripheral give the microcontroller a convenient method of control. Unfortunately, driver modules have several drawbacks. They can be pricey, they are somewhat specialized and their output transistors generally have a higher RDS(on) rating (MOSFET on resistance) than discrete transistors.
If we take a look at things from a mechatronics' point of view, one of the first questions to ask is, do we really need the integrated driver or can we move the control logic into another section of the design and lower the cost? Fortunately, it turns out that microcontroller manufacturers have built H-bridge drive-control functions into their PWM peripherals for the last couple of years.
The PWM peripheral usually includes not only ½ and full H-bridge drive outputs, but they also often include a comparator-based shutdown feature (see Figure 2, above). This H-bridge drive function means a PWM peripheral can directly drive four discrete transistors in an H-bridge, eliminating the need for the control functions present in an integrated driver.
The comparator-based shutdown can then be used for either an over-current shutdown or cycle-by-cycle current limiting for torque control. In some newer PWM peripherals, the comparators also include a Digital-to-Analog Converter (DAC) for controlling the comparator threshold, allowing both speed and torque control in the motor drive. The only external components required are the MOSFET transistors and simple level — translating MOSFET gate drivers.
Sensors are an area where a mechatronics' approach can really cut costs out of a design, because it allows designers to integrate their sensors directly into the system mechanics. For example, consider an angle sensor used to monitor the output of a servo control (see Figure 3, page M15). A typical design would couple either an optical encoder or potentiometer to the shaft and then monitor the output with a counter or ADC.
The cost challenge in this design is that it now requires a shaft coupler, sensor and mounting brackets, and increases the material costs and assembly time. If we go with the optical-encoder option, we also add a home-limit switch and the circuitry necessary to interface the encoder to our counters.
A mechatronics' view of the design might ask, why are we adding a sensor to the existing shaft? Why not instead build the sensor into the shaft? Additionally, if we want an absolute angle, why are we using a dead-reckoning system? Why not measure the angle directly?
These are all are good questions. Is there some way we can build the sensor into the shaft and get an absolute position? The answer is probably.
Let us start with a list of the following sensor requirements:
Measure of the absolute angle
Minimal drag on the shaft
Looking at the problem from an electrical and mechanical point of view, we have several options:
An optical system with a gray scale film
An inductive system, such as a resolver
A capacitive system
From a reliability point of view, we'll eliminate the potentiometer at the start. A potentiometer's reliance on a wiper limits its lifetime, due to wear.
The optical option may work if we have a relatively clean environment, but dust and lubricant can be problematic for this option.
An inductive system is not affected by dust and lubricant, but the complexity of mounting a sense coil on the shaft, along with brushes to carry the signal, gets us back to dust and wear problems.
A capacitive sensor is not affected by dust or lubricant and it doesn't have to make contact with the shaft, other than to ground it. There is no wear issue and the only addition to the shaft is a half-circle plate of something conductive. The capacitor's second plate is another conductor fixed to the frame of the system and held close to the one on the shaft (see Figure 3, left). Measurement of the capacitance can be done with one of the several capacitive-touch blocks currently out on the market, or by using a simple voltage comparator-based system to make a charge/discharge-based ADC or relaxation.
The result is a relatively simple sensor that can be measured without actually touching the shaft. The conversion to a digital number can be accomplished using some simple, on-chip microcontroller peripherals. Additionally, this type of sensor is relatively immune to dust and lubricants. More importantly, it is inexpensive.
In each of these examples, all we have really done is bypass the typical building-block solution in favor of a more system-based approach. This approach takes the entire system into consideration and allocates functions based upon which section can accomplish the work in the most cost-effective way, while retaining the integrity and accuracy of the system. In the end, this is what a mechatronics' design approach is — evaluate the entire system and use a design that makes sense for the complete system, rather than just the individual blocks.