Chapter 1: Introduction and Research Foundation
In the contemporary context of escalating energy scarcity and environmental degradation, the global community has recognized the critical importance of energy conservation and emission reduction. Pure electric vehicles have emerged as an effective solution for green transportation due to their distinctive characteristics of zero tailpipe emissions, exceptional energy efficiency, minimal operational noise, and reduced maintenance costs when compared to conventional vehicles powered by internal combustion engines. The global sales volume of pure electric vehicles has demonstrated a consistent upward trajectory over recent years, with remarkable market penetration rates being witnessed across international markets.
Despite the increasing market share of pure electric vehicles, several technical challenges continue to impede their widespread adoption, including limitations in driving range, suboptimal power performance during uphill driving conditions, and areas requiring refinement in ride comfort. The optimization of transmission systems has been identified as a crucial pathway for enhancing these aforementioned performance metrics. While conventional single-speed reduction gearboxes offer the benefits of structural simplicity and cost-effectiveness, their fixed gear ratios constrain the operational range where the electric motor can achieve high efficiency, thereby limiting the vehicle’s ability to adapt to varying driver demands for dynamic performance or optimal energy consumption. Consequently, multi-speed Automated Mechanical Transmissions (AMT) have gained prominence as the preferred choice for pure electric vehicles.
In this thesis, a systematic approach focusing on a heavy-duty pure electric vehicle equipped with a two-speed AMT was adopted, with particular emphasis on the development of an adaptive shift control strategy intended to improve the vehicle’s overall performance indices including dynamic responsiveness, energy efficiency, and ride comfort. Within the framework of pure electric vehicles, the significance of a properly designed shift control strategy cannot be overstated, because the transmission system acts as a vital interface between the electric machine and the driving wheels, fundamentally governing how effectively the powertrain’s capabilities are exploited under varying operational scenarios. A substantial body of research by scholars worldwide has investigated diverse facets of two-speed AMT technology for pure electric vehicles, encompassing the areas of shift strategy development through intelligent algorithms, powertrain parameter matching, comprehensive energy management frameworks, and advanced simulation methodologies. These investigations fundamentally differ in their technical emphasis, with some researchers prioritizing the economic efficiency or longitudinal dynamic quality of pure electric vehicles, while others place greater emphasis on the real-time responsiveness and adaptability of the control algorithms under uncertain driving conditions.
An examination of the scholarly landscape reveals that two-speed AMT systems for pure electric vehicles can significantly enhance the operational efficiency of the traction motor by enabling the motor to function more frequently within its designated high-efficiency map zone. This is achieved through the deliberate shifting between a lower gear ratio – which amplifies the motor’s output torque at the expense of higher speeds and power consumption – and a higher gear ratio that reduces motor speed for improved efficiency when the vehicle is cruising at high velocities. From the standpoint of a control system designer, however, the two-speed AMT introduces substantial complexity due to the need to execute rapid and seamless gear engagements that minimize driveline torque interruptions, transient shocks, and component wear. In response to these inherent challenges, engineers have sought to design sophisticated control strategies that account not only for the present operating point but also for the predicted future operating trajectory of the vehicle and the preferences of the human operator.
Chapter 2: Dynamic Modeling and Simulation Model Construction
In this chapter, I meticulously carried out the structural design and dynamic analysis of the two-speed AMT mechanism intended for pure electric vehicles by establishing a comprehensive simulation framework in the MATLAB/Simulink software environment. This framework incorporates high-fidelity models of key subsystems to provide a reliable virtual test bed for control strategy evaluation.
2.1 Selection of the Gearshift Actuator
For the execution of gear shifts within the transmission under study, I have rationally selected the locking-ring synchronizer as the primary functional and structural element. The synchronizer is an integral mechanical component within the two-speed AMT that ensures that the rotational speeds of the input shaft and the target output gear become sufficiently aligned before positive engagement is attempted, thereby mitigating the unappealing noise, severe wear, and structural fatigue that would otherwise arise from clashing gear teeth during high-speed meshing. In the context of developing a control-oriented model, I based my work on the functional logic of the gear shift mechanism, as shown in the experimental setup for the shifting mechanism in this research. The mechanical system comprises the synchronizer ring, the hub sleeve (or engagement sleeve), sliding keys, spring retainers, and the conical friction surfaces of the target gears. The force exerted by the shift actuator advances the sliding sleeve, which drives the keys against the synchronizer ring. Friction between the conical surfaces of the ring and the gear creates a torque that rapidly eliminates the angular velocity difference between them. This friction-induced synchronization process is fundamentally important for achieving high shift quality.
Figure illustrates the five distinct and sequential stages of engagement through which the synchronizer transitions during a complete gearshift event, from the initial operating state to the final positive locking of the gear:
(a) The pre-synchronization stage, in which the synchronizer ring is moved into initial contact with the target gear cone, thereby beginning to generate a friction torque.
(b) The synchronization phase, during which the generated friction torque actively brings the rotational speeds of the driving and driven members into conformity.
(c) The lock-release stage, initiated after the synchronization process has been successfully completed.
(d) The meshing stage, where the sliding sleeve is displaced across the synchronizer ring to engage with the gear’s engagement teeth.
(e) The post-synchronization state, culminating in the complete mental alignment of the gear and the establishment of a solid path for torque transmission capability.
2.2 Dynamics Analysis of the Shift Actuator
To ensure that the developed shifting laws and actuator control algorithms can achieve the desired outcomes, it is essential to subject the physical shift actuator system to a rigorous dynamic analysis. This provides the foundational theoretical reasoning and parameter estimation required for subsequent control system tuning. In particular, the shift smoothness of a two-speed AMT is a multi-faceted concept. In this thesis, I have selected a total of five key evaluation indices, which together provide a comprehensive and quantitative overview, namely the synchronization force at the actuator, the gearshift impulse, the vehicle jerk, the friction work of the synchronization process, and the duration of driveline power interruption. These indices serve to rigorously compare the performance of the newly formulated optimized strategy against that of a conventional gearshift strategy.
The determination of the moments of inertia of the relevant rotating components is of vital importance for the calculation of synchronization characteristics. For example, the moment of inertia (\(J_1\)) of the input shaft, the moment of inertia with the synchronizer shaft (\(J_2\)), and the moment of inertia of the output shaft (\(J_3\)) were collected as core examples. In this thesis, I completed an analysis of the driveline parameters using data from a representative heavy-duty powertrain. For the low-speed gear, the total equivalent moment of inertia that must be synchronized (\(J_{\text{low}}\)) was computed to be approximately \(0.4276 \text{ kg·m}^2\). For the high-speed gear, the equivalent moment of inertia (\(J_{\text{high}}\)) was found to be \(0.4663 \text{ kg·m}^2\).
The calculation of the shift force differs based on the stage of the shift process. For instance, the force required to disengage the self-locking mechanism during the pre-synchronization stage is approximated by examining the resistance force of the self-locking device:
$$f_{sk} = f_k (2 \sin\alpha + \mu_k \cos\alpha)$$
$$F_H = f_{sk}$$
where \(f_k\) is the pre-tightening force of the interlock, \(\mu_k\) is the coefficient of friction between the sleeve and the sliding key, \(\alpha\) is the contact angle, and \(F_H\) is the shift force. During the main synchronization phase, the synchronization torque (\(T_C\)) and the required axial force (\(F\)) have a deterministic relationship, governed by the cone angle \(\phi\) and the mean cone radius \(R_C\):
$$F = \frac{T_C \sin\phi}{\mu_c R_C}$$
$$T_C = J_R \frac{\Delta\omega}{t}$$
By combining the above equations, the time required for synchronization can be derived:
$$t = \frac{J_R \Delta\omega \sin\phi}{\mu_c R_C F}$$
In these equations, \(\mu_c\) represents the coefficient of friction between the cone surfaces, \(J_R\) is the rotational inertia of the components attached to the synchronizer’s input side, and \(\Delta\omega\) represents the change in angular velocity. When analyzing shift smoothness, the vehicle jerk is defined as the derivative of the vehicle’s longitudinal acceleration. I calculated the vehicle impact using the formula:
$$j = \frac{d^2 v}{dt^2}$$
Similarly, the frictional work (\(W\)) that results from the relative slip during synchronization, which is directly related to heat generation and wear, and the shift impulse (\(I\)), which is evaluated as the integral of the shift force over time, were calculated with the following formulas:
$$W = \int_{t_0}^{t_1} T_C \Delta\omega \, dt$$
$$I = \int_{0}^{t} F_H \, dt$$
2.3 Development of the Vehicle and Subsystem Models
In order to simulate and subsequently control the vehicle dynamics, I created models for each major component of the electric drive system.
1. Traction Motor Model. This study adopts a permanent magnet synchronous motor (PMSM) as the traction motor for the pure electric vehicle. PMSM motors are favored in the automotive industry because of their high power density, superior efficiency, and excellent dynamic response. In the synchronous rotating \(d-q\) reference frame, the stator voltage equations are:
$$u_d = R_s i_d + L_d \frac{di_d}{dt} – \omega_e L_q i_q$$
$$u_q = R_s i_q + L_q \frac{di_q}{dt} + \omega_e (L_d i_d + \psi_f)$$
with \(R_s\) being the stator resistance, \(L_d\) and \(L_q\) the inductances, \(\psi_f\) the permanent magnet flux linkage, and \(\omega_e\) the electrical angular velocity. The corresponding electromagnetic torque (\(\tau_e\)) is proportional to the currents:
$$\tau_e = 1.5 p [\psi_f i_q + (L_d – L_q) i_d i_q]$$
where \(p\) is the number of pole pairs. In this thesis the maximum torque and rated power of this motor are taken from the delivered vehicle hardware system: the maximum speed typically exceeds 8000 RPM and the peak torque is as high as 1100 N·m, with the rated power exceeding 165 kW.
2. High-Voltage Battery Model. The power battery pack was modeled using a combination of circuit theory and empirical relationships that relate the open-circuit voltage and internal resistance to the battery’s state of charge (SOC):
$$E_{SOC} = E_0 + \sum_{i=1}^{5} E_i SOC^i$$
$$R_{SOC} = R_0 \left( \delta_0 + \sum_{i=1}^{6} \lambda_i SOC^i \right)$$
In this model, \(E_{SOC}\) and \(R_{SOC}\) represent the electromotive force and internal resistance at the given state of charge, respectively, while \(E_i\) and \(\lambda_i\) are polynomial fitting coefficients used to capture the nonlinear relationship. The SOC is typically updated via coulomb counting:
$$SOC(t+1) = SOC(t) – \frac{I_{bat}(t)}{Q_{bat}}$$
where \(I_{bat}\) is the current and \(Q_{bat}\) is the nominal capacity. The battery model calculates the instantaneous current based on the power demand of the motor, and it allows for the potential to recover energy, i.e., when the vehicle is braking, the energy is fed back to the battery through the traction motor acting as a generator.
3. Drivetrain Model. The drivetrain model in the thesis contains the main reducer, the final drive, and the transmission shafts. The gear ratios in the system are critical parameters for the control logic. In this model, the relation between vehicle speed and motor speed was derived within the context of the numerical values of the transmission gear ratio \(i_1=2.05\) and \(i_2=1.03\), as well as the final drive ratio \(i_0\). The implementation of the following relationship between the rotational speed of the vehicle wheels and the vehicle speed is fundamental to driver acceleration prediction:
$$v = \frac{2\pi r n_0}{60 i_0 i_g}$$
where \(i_g\) is the gear ratio (\(i_1\) or \(i_2\)), \(n_0\) is the wheel speed, and \(r\) is the rolling radius of the vehicle tires.
4. Vehicle Longitudinal Dynamics. The longitudinal dynamics of the pure electric vehicle were derived from the balance of tractive forces and resistances:
$$F_t = F_f + F_w + F_j + F_i$$
The tractive force \(F_t\) is computed as \(F_t = \frac{T_e \eta i_g i_0}{r}\), where \(\eta\) is the driveline efficiency. The rolling resistance is \(F_f = C_f mg \cos\theta\), the aerodynamic drag is \(F_w = \frac{1}{2} \rho C_d A v^2\), the acceleration resistance is \(F_j = m \delta \frac{dv}{dt}\), and the gradient resistance due to the slope is \(F_i = mg \sin\theta\).
2.4 Verification of the Baseline Shift Control Model
Before studying the advanced control strategy, I built and verified a baseline system that operates using a conventional speed-based gearshift logic. This baseline control model provides a standard for evaluating the improvements obtained through the optimization process. The model block diagram of this baseline system, implemented within the MATLAB/Simulink environment, was used to simulate the vehicle’s reaction to the INRETS driving cycle. The simulation results show that the vehicle can follow the desired driving cycle with good precision. The gear shifting pattern is consistent with common sense for a two-speed vehicle, and the battery SOC variations during the entire cycle of operation provide concrete verification of the robustness of the pure electric vehicle drivetrain model, thereby laying the foundation for simulation-based research into advanced control policies.
Chapter 3: Research on Shift Control Strategy of Electric Vehicles based on Driving Style and Vehicle Speed Prediction
In light of the identified weaknesses of the traditional speed-based shift control alone for pure electric vehicles, this chapter presents a comprehensive study focused on adaptive shift control: The strategy includes recognizing the driving style of the current driver, predicting the future course of the speed trajectory, and incorporating fuzzy logic control to adapt the gearshift decision-making process to the specific operational environment detected in real time.
3.1 Introduction to Strategy Design
The optimization strategy adopted in this paper is specifically tailored to a two-speed AMT, so as to achieve significant improvements in dynamic performance, energy efficiency, and driving comfort. The core idea is to formulate the shift control logic using two distinct control modules. The first module, the gear decision model, focuses on establishing an effective “when to shift” logic. The second module, the synchronizer shift control model, addresses the “how to shift” manner. In the design of the gear decision model, the topics are all about accurately identifying the driver’s needs. A driving style recognition model identifies the individual behavior as conservative, standard, or aggressive based on dynamic vehicle feedback. A velocity prediction model conjectures the upcoming traffic situation. The overall optimization strategy is based on the observation that a vehicle’s jerk, which is the derivative of acceleration, directly correlates with the intensity of use of the accelerator pedal, as shown in the following relationship, which connects the tractive jerk to the torque gradient of the traction motor:
$$\frac{dT_e}{dt} = \frac{(I_{out} + i_g I_{in}) J}{i_g i_0 / r}$$
3.2 Construction of the Driving Style Recognition Model based on Jerk
Since drivers show significant differences in their operating habits, designing a shift schedule that takes only a single standard into consideration inevitably fails to provide optimal satisfaction to all users. As part of the intelligent control architecture developed here, a model was constructed to classify a driver’s style based exclusively on input from the vehicle’s longitudinal movement behavior. I chose the vehicle jerk as the representative variable, as it directly embodies driver behavior and is influenced by the environment. The vehicle acceleration signal is processed over a time interval of 200 seconds. To make the recognition algorithm robust against noise and offset errors, a jerk analysis coefficient, \(R_{\text{driver}}\), is introduced to characterize the dynamic shift patterns. This coefficient is computed as the ratio of the standard deviation of the jerk signal to its absolute mean value:
$$R_{\text{driver}} = \frac{\sqrt{\frac{1}{T} \sum_{i=1}^{T} (J_i – \overline{J})^2}}{\frac{1}{T} \sum_{i=1}^{T} |J_i|}$$
The jerk itself is calculated from the acceleration derivative, \(J_i = (a_{i+1} – a_i)/\Delta t\). The result is then classified using empirical thresholds. When the calculated coefficient \(R_{\text{driver}}\) falls below a specified level of 0.3, the driver is categorized as conservative (normal). When the coefficient exceeds another higher threshold, for example 0.8, the driver is considered aggressive (dynamic). Intermediate values are designated as standard. This identifies the appropriate driver behavior mode.
3.3 Vehicle Speed Prediction Model based on Markov Chains and Its Validation
To anticipate future vehicle speed changes precisely, I formulated a velocity predictor based on Markov chain theory. Markov chain prediction is a classical method often used in trajectory and speed prediction applications because it does not rely on a complex deterministic physical law. The chain is defined using a discretized representation of the vehicle speed and acceleration. The available vehicle speed range was discretized into a finite set corresponding to 30 states, and the acceleration values were similarly discretized into 30 sets. The fundamental element of the predictor is the probability transition matrix, which describes the likelihood of achieving a target acceleration based on the current speed level. It is calculated from some sample dataset of historical driving data.
Instead of using just a single step, I developed a four-stage Markov chain model to achieve a longer prediction horizon of 2 seconds. The prediction procedure is visualized and explained. First, the current speed is fed to the first-stage probability transition matrix to identify the most likely value of the acceleration over the next 0.5 seconds. By applying the formula \( v(t + 0.5) = v(t) + a_{t \to t+0.5} \times 0.5 \), one can compute the predicted velocity. This newly estimated speed is then used as input to the second-stage transition matrix to obtain the acceleration \(a_{t+0.5 \to t+1}\) and predict the velocity at the 1s horizon. The process is repeated, using the second-stage probability transition matrix, for forecasts at the 1.5s and 2s time horizons. As the model has four separate stages, effectively equivalent calculations are performed with a 0.5s temporal step to provide more detailed resolution of future changes, which enables more precise control.
To create the state transition probability matrix for the specific model, I used a comprehensive set of real-world driving cycles as sample data. This set includes well-known driving schedules like the UDDS, NYCC, MANHATTAN, HWFET, US06_HWY, WVUCITY, and others available in standard simulation packages. These are diverse driving cycles that span a good variety of traffic conditions. For complete compliance with the general rules, all the sample driving cycles were used to form the final Markov matrix.
To validate the accuracy of the model, the very same cycles were fed back into the prediction model. The model was run for the long-duration sequence formed by the concatenation of 10 typical driving cycles. A comparative plot showing the reference and predicted velocity curves were drawn for the entire duration of 10198 seconds. The results of the analysis indicate a strong alignment of the “predicted operating condition” curve with the “cycle” curve under most of the conditions. The prediction error remained below 3% during steady-state operations, and the predicted acceleration curve also agrees well with the actual curve. This verifies that the chosen approach has good capability of anticipating the vehicle’s future cruise speed.
After validation, the predictor was tested on performance under a new dataset that was not a part of the model-building process, using the LA92 driving cycle. The prediction error remained low, specifically under 5% in steady-state driving, when compared to the actual cycle speed curve. The LA92, Ftp72, and Japan_urban cycles have also produced similar high prediction precision when the vehicle runs on a prediction provided by the Markov chain model.
3.4 Design of the Fuzzy Logic Gear Decision System
The next step in the gear decision model is the design of a fuzzy control system which is able to incorporate the two central pieces of information that have been previously acquired: the recognized driving style and the predicted operation cycle. The control logic considers that both future and current driving conditions and driver characteristics are relevant to select the best gear, but that their relationship cannot be easily described by simple mathematical equations. Thus, fuzzy control appears to be well suited for this task. In my design, a fuzzy controller, identified later as a block within the vehicle model, is responsible for computing an output named the “gear ratio coefficient”. The inputs to this controller are the upcoming “predicted driving conditions”, and the output is given as the current “driving style”. These inputs become normalized by the fuzzy reasoning strategy.
Fuzzy membership functions were created to express the level of each input. For the driving style, I defined three categories using conventional fuzzy sets: 1 (conservative), 2 (normal), and 3 (aggressive). For the future speed, the universe of discourse covers an interval from 0 to 100 km/h, and this is partitioned into five semantic classes: (Very Low, Low, Medium, High, Very High). Likewise, the gear coefficient, representing the target gear number for a two-speed gearbox, has a range from 1 to 2, corresponding to the first gear and second gear. Since the vehicle is a heavy-duty truck and is normally in its top gear, the gear ratio coefficient dimension is divided into multiple fuzzy sets between 1 and 2. The output of the controller is computed using a carefully crafted rule base. This inference engine is based on expert knowledge about matching gear to vehicle operation. For instance, a combination of high predicted speeds and a moderate driving style might imply shifting up to second gear. By contrast, at low vehicle speeds, regardless of the driver’s attitude, a lower gear may be required, potentially more urgently if the driver style is aggressive.
3.5 Design of the Fuzzy PID Shift Actuator Controller
Beyond optimizing the point in time and opportunity for shifting gears, the methodology I devised also ensures optimal performance during the shift action itself. To improve the actual shift dynamics of the synchronizer sleeve so that it matches driver and road conditions, the control system monitors both the rotational speed discrepancy between the two sides of a synchronizer and the rate of change of this speed. As was previously described, this change rate is directly related to the torque dynamics of the traction motor and, through the drive shaft, to the driver’s throttle commands. An aggressive driver, who demands fast shifts, will produce high rates of change in rotational speed differences. A conservative driver operating in typical situations, with slower and smoother speed variations, can be better served by softer and more cautious shift operation to minimize any perturbations in driveline torque and improve comfort.
The speed difference (in units of revolutions per minute) in the synchronizer was discretized into nine fuzzy sets named (very small, small, etc.) across the range from 0 to 10,000 RPM. The rate-of-change of the speed difference over a time step of 0.002s was allotted a universe of discourse up to 5,000,000. The output is a modulation factor called the “shifting motor speed demand coefficient”, which operates over the range from 0.5 to 1. This factor multiplies a nominal shift motor speed signal. The system creates an input value equal to the product of a nominal speed and the fuzzy output to modulate the shift motor speed. A fuzzy control rule table was built based on vehicle dynamics.
The designed fuzzy logic controller is responsible for calculating the demand coefficient. The synchronizer position is measured and fed back via a conventional PID controller. To achieve optimal performance, the speed of the shifting motor is computed, integrated into the position, and compared with the desired setpoint to provide closed-loop control. This architecture enables smooth adaptation of the shifting process without requiring a full model in the main control path. It improves the comfort of the shift action and brings the actual trajectory of the shift actuator into agreement with the established overall control strategy.
Chapter 4: Simulation Results and Experimental Testing of the Gearshift Motor
4.1 Overview of the Integrated Simulation Model
Following the design of the separate control modules described previously, I integrated all the components into a comprehensive simulation model in Simulink. The overall model, depicted in the form of block diagrams, comprises the new gear decision module, the actuator shift module, as well as the standard model of the pure electric vehicle. The gear decision model is diagrammatically shown in the form of the appropriate vehicle structure. The gear decision fuzzy controller receives its inputs and generates a gear command. A critical feature called the adaptive shift blocking mechanism was developed to address the common problem of excessive gear hunting. Field experience has shown that, without intervention, many production two-speed pure electric vehicles repeatedly cycle between gears during slightly varying driving conditions. The newly designed algorithm stores the exact time of the last gearshift and then computes the interval until the next one, yielding a temporary time stamp.
Based on the duration of the time interval, the signal that was originally produced by the fuzzy controller is interpreted in the following logical manner:
(a) If the output of the gear coefficient exceeds 1.8 while the motor speed is approaching its maximum limit, the target gear is set to second gear.
(b) If the output coefficient is above 1.8 and more than 60 seconds have passed since the last shift occurred, the target gear will switch to the second gear.
(c) Should the coefficient remain above 1.2 but lower than 1.8, the target gear remains unchanged.
(d) When the coefficient is less than 1.2 and more than 60 seconds have elapsed, the target gear will shift to first gear.
This time-based restraint in shifting yields a stable drive state and avoids oscillatory, hunting shifts.
4.2 Simulation Results for Speed Prediction and Driving Cycle Performance
The 4-stage Markov chain predictor was first subjected to rigorous verification on the same 10 cycles used in its design stage, and it demonstrated a prediction error that remained less than 3% for the whole duration. This high level of accuracy confirmed the predictor’s fidelity in matching vehicle responses and its applicability for online use.
In the next stage of testing, predictors were executed in unknown traffic scenarios: LA92, Ftp72, and Japan_urban. The comparison plots of the actual and predicted speed profiles in these test cycles are shown later. In these typical city and highway driving conditions, the vehicle speed follows the predicted path closely for stable periods, and prediction errors generally remain below 5%.
4.3 Frequency of Gear Changes Analysis
I compared the performance of optimized and traditional gear shift control approaches in terms of gearshift frequency. Table 1 summarizes the results for different cycles and driver styles. Using the LA92 driving cycle as the reference condition, the number of shifts executed by the vehicle was assessed for traditional speed-based logic and for our adaptive logic under three distinct driver behavioral modes.
| Driver Style / Cycle | Control Strategy | Number of Gear Changes | Reduction Percentage |
|---|---|---|---|
| Conservative (LA92) | Traditional Strategy | 42 | – |
| Conservative (LA92) | Optimized Strategy | 18 | 57.14% |
| Standard (LA92) | Optimized Strategy | 16 | 61.90% |
| Aggressive (LA92) | Optimized Strategy | 22 | 47.62% |
| Conservative (Ftp72) | Traditional Strategy | 36 | – |
| Conservative (Ftp72) | Optimized Strategy | 23 | 36.11% |
| Conservative (Japan_urban) | Traditional Strategy | 36 | – |
| Conservative (Japan_urban) | Optimized Strategy | 22 | 38.89% |
From this table, it is clear that there is a marked reduction in shift frequency while driving with the presented control strategy when compared to the conventional baseline. Indeed, in all examined scenarios it was possible to significantly reduce the number of gear shifts for a given driving cycle. This reduction in unnecessary gear changes helps improve ride comfort and lower mechanical system wear, without losing performance. In contrast to the standard state-flow control, my control strategy proves to properly anticipate the most appropriate gear based on driver style and future traffic predictions, resulting in fewer hunting shifts.
4.4 Analysis of Shift Dynamics and Powertrain Performance
The dynamic behavior of a single gear shift event is analyzed to better understand the physical interactions occurring within the transmission, particularly along the path of the actuator position. The data in the following paragraphs describes a period of 8 seconds that contains a start from standstill, an upshift from low gear to the high gear, and an eventual downshift to the low gear. The gear shift position signal progresses over time. When the y-axis is zero, the gear is positioned in neutral. A negative change in the gear shift position means that the synchronizer is being pushed towards the low-gear position. The engagement to the low gear occurs with a total shift time of 0.8s when the position has reached its limit of -12.5mm. During the upshift, a positive displacement of the shift actuator occurs, and because the speed difference between the two sides of the synchronizer initially exceeds 1500 rpm, there is an initial waiting period while the speeds are synchronized. After this initial synchronization period of 0.22 seconds, the sleeve again traverses toward the high gear in 1.76s, where the position reaches +12.5 mm for the high gear engagement. When downshifting from a higher speed to a lower gear happens, the speed difference becomes less than 1500 rpm, allowing rapid synchronization throughout the entire shift sequence over a period of 1.56 seconds.
The speed differential values between the driving and the driven plates exhibit high values when the vehicle is in motion and the target and current gear speeds are unequal, especially if shifts need to be synchronized at higher speeds. The analysis of the shift actuator or shifting fork angle signal shows that it follows the shift actuator position signal faithfully, with its oscillation within the range from -7 to 7 degrees as seen in the data plots. The structural dependencies verified by this data validate the correct operation of the synchronization actuator model under these controller signals.
To quantitatively examine the performance of the vehicle in longitudinal dynamics and efficiency using the proposed methodology, a series of simulations was run comparing the new control strategy with the traditional speed-based gear shifting method. Vehicle dynamic performance is characterized by the traction motor torque. Metrics have been drawn as cumulative values over the test run. Table 2 summarizes the simulation results of the drivetrain performance metrics before and after optimizing the shift control model.
| Metric | Case Study | Baseline | Optimized | Change |
|---|---|---|---|---|
| Vehicle Driving Force (N·m·h) | LA92 | 14.46 | 17.55 | +21.36% |
| Vehicle Driving Force (N·m·h) | Ftp72 | 20.51 | 21.97 | +7.18% |
| Vehicle Driving Force (N·m·h) | Japan_urban | 15.67 | 16.10 | +2.74% |
| Energy Consumption (kW·h) | LA92 | 5.97 | 5.73 | -4.02% |
| Energy Consumption (kW·h) | Ftp72 | 8.94 | 8.61 | -3.69% |
| Energy Consumption (kW·h) | Japan_urban | 7.18 | 6.97 | -2.92% |
| Driving Range Extension | LA92 | Baseline | – | +3.94% |
| Driving Range Extension | Ftp72 | Baseline | – | +3.70% |
| Driving Range Extension | Japan_urban | Baseline | – | +2.94% |
| Total Power Interruption Time | LA92 | 47.66s | 21.24s | -55.43% |
| Total Power Interruption Time | Ftp72 | 41.02s | 27.22s | -33.64% |
| Total Power Interruption Time | Japan_urban | 40.98s | 25.72s | -37.24% |
The results consistently show a marked improvement in the driving dynamics via an increased vehicle driving effort in all three cycles. At the same time, the vehicle has achieved a lower energy consumption as the designed shifting schedule avoids low-efficiency operation points and predicts the upcoming route, allowing the pure electric vehicle to extend its driving range in each situation. The improvement of driving range is particularly beneficial for heavy-duty electric vehicles, which often suffer from operational range anxiety.
4.5 Evaluation of Gearshift Comfort and Shock Mitigation
The enhanced shifting logic was benchmarked against the conventional shifting logic on the basis of two criteria related to the comfort during the shift event. The shifting force produced by the synchronization actuation system shows time histories. In the LA92 cycle, the baseline controller executes 42 equal peaks within the shift event, whereas the advanced shift controller generates 18 shift force peaks, resulting in a 57.14% reduction in the frequency of force peaks.
To capture the total level of load introduced into the driveline by all shift maneuvers over the whole driving cycle, I integrated the shift impulse over the vehicle travel time. In the LA92 cycle, the impulse-time curve for the baseline control terminates at 66,687 N·s², whereas the optimized control produced a measure of only 45,475 N·s², which is a 31.81% reduction in overall powertrain shock. Comparative analyses in the Ftp72 and Japan_urban drive cycles produced more moderate decreases, with a value close to 0% in Ftp72 and a 4.62% decrease in the Japan_urban test, respectively. The adoption of the forecast of the path condition has provided additional flexibility to lower a certain part of the shift-related roughness.
4.6 Analysis of Driveline Power Interruption Time
The time elapsed when the driveline interrupts its power delivery path due to gear unmeshing and subsequent meshing has a direct effect on the responsiveness felt by the driver. In this thesis, the time of power interruption is assessed as the time delay between the target gear change signal and completion of the shift action. Table 2 shows that in the LA92 case, thanks to the strong reduction in shift count, the overall interruption interval becomes drastically shorter, dropping from 47.66 s to only 21.24 s. In the Ftp72 and Japan_urban schedules, the improvement in total interruption time is equally present with a reduction of 33.64% and 37.24%, respectively. This improvement is due largely to a reduced shift frequency rather than to a reduction of single shift time, whose act may sometimes be lengthened deliberately to smooth the transient motions. Despite a slight average increase in shift time, the total gap is still shorter, which indicates a net overall gain in responsiveness.
4.7 Gearshift Motor Functional Testing
Considering the limitations in laboratory resources, the approach taken in this thesis was to perform an unloaded functional test bench evaluation to validate the feasibility of the proposed new shifting strategy through speed control of the gearshift motor. A dedicated experimental setup was built, in which a dedicated permanent magnet synchronous motor for gear shifting (200W, 24VDC, rated speed 3000RPM) is mechanically connected with a motor controller. A PC with TODE-MotorHost user interface software is connected through a CAN communication link to the inverter. The motor controller receives the demanded speed values sent from the PC according to actual shift commands and transmits to the inverter, which drives the motor. The test setup was operated with the motor disconnected from the gearbox, allowing evaluation of the dynamic motor response to speed commands without significant loads.
The command speed was initially set to 200 RPM in the control software. The motor speed trace around the target response is depicted in the results. The measured speed shows a minor oscillation of 0.51 RPM around this reference. The error between commanded and actual speed is less than 1%. After the setpoint was changed to a value of -30 RPM, thereby reversing the rotation direction of the gearshift motor, the average speed was maintained at 30.00 RPM with a small oscillation amplitude of only 0.25 RPM. The measured speed difference as compared to the desired command, again remains below 1%. In the test scenarios, the motor ran with no failures and delivered exactly the requested speed, in the desired direction.
The results indicate that the functional performance of the experimental motor is reliable, and it confirms that the proposed approach for regulating shift motor speed is promising for implementing active control of shift actuator motion. These experimental results establish the necessary conditions for further development of shift-by-wire systems, which need a high degree of accuracy and repeatability, corresponding to control motor speed and enhancing the shift process in pure electric vehicles.
Chapter 5: Conclusion and Future Outlook
This thesis addressed the research problem of an adaptive shifting strategy for a two-speed AMT installed in pure electric vehicles, while focusing on improving the vehicle efficiency while retaining comfort and dynamic quality. With the goal of devising a control system that can be reliable under both structured and dynamic driving environments, I developed a simulation model for a heavy duty pure electric vehicle powertrain in MATLAB/Simulink. I integrated it with a purposely built optimization architecture. This architecture makes decisions by two main blocks: the gear position determination block and a synchronized shift execution block. In order to optimize the shift decision, a driving style recognition system using longitudinal jerk signal processing was implemented, which classifies driver behavior into conservative, standard, or aggressive categories. In parallel, based on the theory of Markov chains, a four-stage speed prediction logic formed a model to predict the future path over a horizon of two seconds by making use of a pre-calculated acceleration matrix built on several driving cycles. The shift decision system accordingly utilized a fuzzy controller, whose input signals were the driver style and upcoming speed. The fuzzy controller then returned the gear coefficient. Furthermore, an adaptive algorithm processed the fuzzy controller output by taking into account the shift time interval and the current speed of the traction motor. In this way, the gearshift controller is also prevented from unnecessarily cycling between the gears. In parallel, the actuator shift controller was refined by supervising the speed difference in the synchronizer and its time derivative. A fuzzy PID controller output a modulation coefficient influencing the command speed of the shift motor. All these refinements were intended to improve shift comfort, which has been evaluated based on multiple criteria including shift force, shift impulse, jerk, friction work in the clutch, and power interruption time.
The outcomes of the extensive simulation study proved the merit of the proposed approach. The model produced accurate vehicle speed forecasts below 5% error in a steady-state drive in relation to the unknown test cycles. A single test cycle with different styles saw gearshift frequency decrease 57.14% for a conservative driver, 61.90% for a standard style, and 47.62% for an aggressive driver when only the driving style has changed. The shift frequency also dropped substantially in all test cases when the driving cycle conditions were varied. The power of the traction motor increased by 21.36% and consumption was lowered by around 4.02%, leading to an overall 3.94% range extension in the LA92 cycle, which proved to be the most sensitive scenario. For this drive cycle, total time of power interruption was reduced by as much as 55.43%. The experimental program verified that the gearshift motor responds with high accuracy to the command speed (errors <1%), thereby confirming the fundamental possibility of achieving well controlled shifting maneuvers using these principles.
Despite this potential, I also recognize the limitations of the current study. The modeling of the driveline in this work is still comparatively simple and does not cover all energy dissipation routes. In addition, Markov chain prediction could be further enhanced using additional records or deep learning, thereby enabling longer or more accurate prediction horizons that extend to 5 seconds. This research should also be pursued by building a more complete Carsim + Simulink co-simulation environment to better emulate interactions and identify kinematic defects. Eventually, the authors suggest tests of these strategies in full vehicle prototypes to gather real-world experimental data on the effectiveness of the optimization. Such test campaigns can serve to reliably assess benefits of driving style and speed prediction functions in the field. Therefore, the presented pure electric vehicle two-speed AMT shifting controller opens a viable road for the further development of intelligent vehicle drive systems.