In the following, we will briefly outline the most important basics on which our technology is based. We provide further details for experts below.
In order to control three-phase machines dynamically and efficiently using today’s standard methods, a position sensor is necessary for measuring the rotor position. The elimination of this position sensor would offer advantages for almost every application, such as cost reduction, space saving, reduction of failure probability or increase of robustness. It is advisable to avoid these expensive and vulnerable mechanical sensors particularly in cost-sensitive and safety-critical applications as well as under harsh environmental conditions.
Problems of sensorless control schemes
However, common sensorless control schemes lead to high development costs and risks due to their technological requirements, and show some disadvantages compared with control schemes using a position sensor. Examples include the necessary control bandwidth reduction, the audible noise emission at standstill, parameter dependencey, problems affecting the transition between low and high speed and the lack of suitability for machines with concentrated winding.
Depending on the application, some of the points mentioned are usually exclusion criteria for many industry applications. Therefore, injection methods are still almost not available on the market, and the use of position sensors for lower speed range is generally unavoidable.
Aim: Efficiency and performance without position sensor
Bitflux has now succeeded in overcoming the known problems of sensorless control. Through 20 man-years of research and development in the field of sensorless control, Bitflux has a wealth of know-how in the field of drive control. Hence, today Bitflux’ technology allows to realize applications with higher demand on efficiency and performance and, as a result, to reduce system costs and failure probability.
Learn more about the Bitflux technology
According to the founders of Bitflux, one of the most important reasons – maybe even the most important reason – for many possible full-speed applications not having been realized sensorless yet, is the fact that the methods are technologically complex. It takes many man-years of development to study the academic state of the art, to implement and to compare the individual sub-methods and to bring a reasonable combination with good characteristics into serial production. Moreover, the drive manufacturer has to maintain the sophisticated know-how permanently for maintenance and support purposes even after implementation. Hence, this “investment” (development cost and risk) is only worthwhile with correspondingly large numbers of units and excludes a large number of small and medium-sized enterprises from sensorless control. On the other hand, the equally complex know-how for the production of position sensors is “out-sourced” anyway, and is thus available for all companies. After more than 20 man-years of R&D in the field of sensorless control, Bitflux has a comprehensive knowledge of international literature and has also made significant progress in solving a large number of problems in literature (see the following sections). Consequently, the challenge and central development task for Bitflux was to standardize and simplify this highly complex algorithm-package so as to be applicable to any synchronous machine and software-engineer without any sensorless experience. In particular, the commonly large number of tuning parameters in sensorless control were logically combined and externally removed. The result is the sensorless software library “dynAIMx®” with clear interfaces and distinct parameterization based on machine data. DynAIMx® brings sensorless control with the best possible performance available also for medium-sized drive-manufacturers.
Sensorless methods from literature are mostly based on filter or observer structures, which are subject to a general trade-off: the cleaner and quieter the estimated signal should be, the more delayed it becomes. Smoothness is, hence, bought with dynamic reduction. The severity of this conflict depends to a large extent on the signal-to-noise ratio (SNR) of the source signal, which in sensorless control schemes is particularly critical in the low speed range. Since, on the one hand, the EMK is weak (to non-existent) and on the other hand the inductances in general can only be determined with a low signal component (saliency ratio often <10%), a poor SNR is generally present in the low speed range. This relationship is even more critical in acoustically sensitive applications, which require a preferably weak injection. In addition, the stability-critical switchover process from low to high speed methods in literature often only allows a slow transition of the switching area. For all these reasons, sensorless control methods for the entire speed range known from literature are known to be suitable only for applications with low dynamic demands. We have been able to work out a method that combines EMF- and inductance information without reverting to observer or filter structures. In principle, our method always refers to both sources of information and doesn’t switch from one method to the other. As a result, our technology achieves an extraordinarily high control bandwidth for sensorless methods (comparable to resolvers) for the entire speed range and is no longer subject to the usual transition problem or the target conflict between dynamics and injection noise. If, in the higher speed range, the EMF-information is strong enough, the injection can nevertheless be switched off without any losses, in order to become silent and to utilize the full control range.
Good, non-linear fundamental model based methods can theoretically operate with arbitrarily high torque in the high speed range – low-speed methods, by contrast, can’t. Low-speed methods evaluate the magnetic anisotropy (directional dependency of the inductances), which is influenced by the torque-producing current. Whit increasing current the orientation of the anisotropy is increasingly different form the rotor angle. That is why, fFrom 2005 onwards, literature contains the advice to compense this machine-specific shift in software (e.g. by means of a look-up table). However, this correction also only functions up to another, extended torque limit, after which low-speed methods become unstable again. The exact value of this torque or load limit depends on the type of machine, i.e. on the design, and is on the average at approximately two times nominal torques. In good cases, even four times is possible, but in bad cases only half or less. Hence, there are many publications in literature on design guidelines for sensorless control, which additionally feed into the compromise between cost, power and efficiency (etc.) and can therefore only lead to limited success. Our above described approach for controlling all (also difficult) machine types is not subject to the load problem. This means that, on the one hand, the sensorless characteristics do no longer have to be considered in machine design, but the machine can be designed solely according to performance and cost. And on the other hand this means that, former futile applications, such as water-cooled drives in the electro mobility, can now be operated without position sensor.
However, one of the original limitations of sensorless control also remains in the Bitflux technology and will, according to the founders, not be overcome in the foreseeable future: accuracy. If the position calculated without sensor is compared to the signal of a highly accurate position sensor, you find some electrical degrees of deviation (for good parameterization 3-5 °), so-called estimation error. For 3 pole pairs this would correspond to 1-2 mechanical degrees of accuracy, with higher pole pair numbers correspondingly less. The reasons for this are complex. Simplifications in modelling (e.g., switching dead time), measurement errors (e.g., current noise and offset), material (e.g., permanent magnets and sheet) or manufacturing tolerances (e.g., air gap) as well as temperature dependencies (e.g., resistance and transistor characteristics) are some examples. Simply spoken, sensorless control uses a highly complex system, which has been designed and optimized for energy conversion, simultaneously as a measuring device. It is possible to consider the sensorless characteristics during design and to achieve improvements thereby, but one will never achieve the quality of a position sensor, in particular with regard to accuracy, without the energy conversion features being lost. Therefore it is important to know that applications with higher demands on accuracy than described above will not be able to do without a position sensor.
The position dependency of the inductances is additionally evaluated at low speeds and at a standstill, when the EMK-signal is very weak. In the progress, an additional signal with a high frequency is impressed and the resulting current response is used to calculate the inductances. This additional signal, which is required for every low-speed method, produces a high-frequency audible noise – a high tone of the machine at low-speed – which can be disruptive in certain noise-sensitive applications. Bitflux has dealt with this subject for many years and has worked out several ways to reduce this noise. On the one hand, we now hold know-how for consulting during the hardware and early software development phases in order to positively influence current measurement and AD conversion (depending on the power range and price segment) and, if possible, also power electronics and motor design. As a result, it is possible to lower the noise significantly or even cancel it completely. On the other hand, we worked out an algorithmically optimal way of evaluating the inductances and integrated this into dynAIMx®, which allows us to reduce the injection and consequently the noise even more without any compromises regarding dynamics (as common for other methods). For instance, 30mA HF current is sufficient for dynAIMx® using a good hardware design with a + -25A current measuring range to calculate the position in a low speed range and to provide an estimation signal with resolver-like dynamics.
The derivation of all low-speed methods known in literature is based on the assumption that the magnetic anisotropy is aligned with the rotor angle and is rotating fixed with it. This means that when the rotor is moving, sinusoidal changes of the phase inductances arise, the anisotropic angle must be corrected by its load-dependent displacement at the worst (see load problem) and can then be used as a rotor-position estimate. However, actually even in supposedly sinusoidal machines there is a slightly harmonic behavior of the anisotropy, which intensifies with increasing torque. Moreover, the trend in modern machine design is to develop machines with concentrated windings because they achieve higher torque and power densities at lower manufacturing cost. Their disadvantage (for sensorless control) is that they do not have sinusoidal inductance curves, but strong harmonics, which leads to local stagnation or even reversing of the anisotropy angle at constant rotor motion. This behavior of the anisotropy makes it no longer possible for any known method to calculate the rotor position at low speed, such that these modern machines are generally considered to be not observable without position sensor. A basic problem of sensorless control is thus its lacking generality – that it is not applicable to all machines, but that each machine type is to be tested for its “suitability”.
We have developed an innovative approach that overcomes the generality problem. This approach regards magnetic anisotropy from a completely different perspective, in which the assignment problem (reversing anisotropy) no longer exists for any machine. Now motors, which have previously been classified as non-observable, are controllable sensorless, whereby the previous test for suitability of a machine does now belong to the past.
In literature, sensorless control methods are usually developed and proposed without regard to the entailed computiational load. Almost all procedures require a variety of filter operations, trigonometry and angle calculations. If, however, a position sensor is to be dispensed with for reasons of cost, additional cost for a larger processor would also be detrimental. Ideally, sensorless algorithms therefore do not lead to any significant additional load of the controller. For this reason, Bitflux developes its algorithms at any time with regard to controller cycles and memory requirements, and structures and optimizes the entire dynAIMx® library with regard to these aspects. DynAIMx®, for example, only uses complex mathematical operations such as sqrt() or atan2(), which were then implemented with only the necessary precision, in indispensable cases. As a result, only a fraction of the computation cycles of the respective fast-math functions is needed. In addition, Bitflux has developed a new injection method called “Square-Injection” based on a square-shaped injection pattern. The intenton was to replace the usual transformations and trigonometric operations by a signed swap of vector components – wich already represents a significant simplification. Surprisingly, all equations finally collapse and the overall anisotropy calculation is reduced to six subtractions – a fraction of the usual computational effort of other anisotropy-based methods.