The Permanent Magnet Synchronous Motor (PMSM) is widely used in industrial drives, new energy vehicles, household appliances, and other fields due to its high efficiency, high power density, and excellent dynamic performance. Traditional PMSM control requires installing position sensors (such as encoders or resolvers) to obtain rotor position information. However, sensors increase system cost, size, and failure rates. Sensorless vector control technology, which estimates rotor position and speed through algorithms, has become a current research hotspot. This article provides an in-depth analysis of the principles, implementation methods, and latest advancements of this technology.
The essence of sensorless technology is to estimate the rotor position in real-time by utilizing measurable motor signals like terminal voltage and current, combined with the mathematical model of the motor. Its theoretical foundations can be categorized into three types:
1. Back-EMF Based Methods: These rely on the characteristic that the back electromotive force (back-EMF) generated by the permanent magnets is related to the rotor position. When the motor operates at medium to high speeds (typically >5% of rated speed), the back-EMF signal is significant. Position information can be extracted using a Phase-Locked Loop (PLL) or a Sliding Mode Observer (SMO). For example, the Extended Electromotive Force (EEMF) model transforms cross-coupling terms between the d and q axes into an extended back-EMF through coordinate transformation, improving estimation accuracy in the medium to high-speed range.
2. High-Frequency Signal Injection Methods: Suitable for zero and low-speed scenarios. A high-frequency voltage signal (e.g., 1-2 kHz sine or square wave) is injected into the stator windings. By utilizing the difference in inductance saliency caused by magnetic saturation effects in the motor, the position is obtained by demodulating the response current. Common methods include rotating high-frequency injection and pulsating high-frequency injection. Technical literature indicates that position error can be controlled within ±5 electrical degrees.
3. Flux-Linkage Observation Methods: Position is calculated by integrating the stator voltage equation to obtain flux linkage and then combining it with the current model. This method requires solving the issue of integral drift, typically addressed using improved pure integrators or combinations with low-pass filters.
> Sliding Mode Observer (SMO): Forces the system state trajectory to converge by constructing a sliding surface, offering strong robustness against parameter disturbances. Technical literature notes that its switching characteristics introduce chattering, which can be optimized using saturation functions or the boundary layer method.
> Adaptive Observers: Such as the Model Reference Adaptive System (MRAS), which uses an equation without position parameters as the reference model and an equation containing parameters as the adjustable model, adjusting the speed estimation value through error feedback.
> Kalman Filter: Incorporates system noise into the state equation, making it suitable for noisy environments, albeit with higher computational load.
Starting from zero speed requires solving the challenge of initial position identification. The high-frequency pulse injection method determines the magnetic pole position by comparing the amplitude of current responses. Technical data indicates that modern algorithms can reduce initial positioning error to within ±10°, meeting the needs of most applications.
High/Low Speed Switching Strategy:
Hybrid control schemes combine high-frequency injection and back-EMF methods: high-frequency injection is used at low speeds, switching to back-EMF observation at medium to high speeds. The key challenge lies in smooth transition. Technical literature mentions a dynamic fusion algorithm based on weighting factors that can avoid switching oscillation.
1. Parameter Sensitivity: Changes in parameters such as resistance and inductance can lead to observation deviations.
Solutions include:
> Online Parameter Identification: Such as Recursive Least Squares (RLS) for real-time motor parameter updates.
> Robust Control Design: H∞ control or fuzzy PID to enhance anti-interference capability.
2. Low-Speed Torque Ripple: High-frequency signal injection may cause additional losses. Employing random frequency modulation or dead-time compensation can mitigate this effect.
3. Deep Flux-Weakening Control: When speed exceeds the base speed, flux weakening is required for speed expansion. Under sensorless control, the voltage equation needs reconstruction, introducing dynamic compensation loops to prevent estimation divergence.
1. New Energy Vehicles: The Tesla Model 3 rear-wheel-drive version uses sensorless Interior PMSM (IPMSM) control, achieving full-speed range position estimation through an Extended Kalman Filter (EKF) algorithm. Eliminating the traditional encoder reduced system weight by 1.5kg.
2. Industrial Servo Systems: Yaskawa's Σ-7 series drives integrate high-frequency injection technology, achieving zero-speed torque control accuracy of ±0.5%.
3. Household Air Conditioners: Gree's third-generation inverter compressor uses a Sliding Mode Observer solution, reducing costs by 20% and passing low-temperature start-up tests at -30°C.
4. Spindle Motor: Achieves a wide range of constant torque, low-speed, and high-torque capability from 1.5 kW to 37 kW through inverter vector control, enabling higher machining efficiency for engraving machines, grinding equipment, and more. A cost-effective air-cooled design ensures low motor temperature rise, eliminating the need for cooling water tanks.
1. AI-Integrated Control: Deep learning applied for observer parameter self-tuning, such as using LSTM networks to predict position errors.
2. Multi-Physics Field Collaborative Observation: Combining non-electrical signals like vibration and noise to enhance reliability.
3. Empowerment by Wide-Bandgap Semiconductors: The high switching frequency of SiC inverters makes high-frequency injection signals easier to extract.
Sensorless vector control technology is transitioning from the laboratory to large-scale applications. With advancements in chip processing power and algorithm optimization, its accuracy and reliability are gradually approaching that of sensored solutions, offering simpler and more economical solutions for motor systems. Future efforts need to further overcome challenges such as ultra-low-speed high-precision control and full-operating-condition parameter self-adaptation to expand its application boundaries into high-end fields like aerospace and precision medical equipment.
