| Multi-level inverters have been widely implemented in rail transit,domestic appliances,aerospace and energy storage systems,and play the key interface roles.For the demand of long,safe,and reliable power supply in above occasions,the improvement of inverter system reliability has become a hot focus of the recent research.Since power devices are relatively vulnerable components in power electronics converters,the repaid diagnosis response and the fault tolerance modulation following diagnosis information can ensure the uninterrupted operation of the system,and avoid the security threat and repair fees due to unwanted shutdown.Taking the T-type three level inverter as the research objective,this thesis explores the characteristics and influence mechanism of power switching device open-circuit faults from various perspectives,and proposes novel schemes and technical support for device state monitoring and fault-tolerant operation of T-type three-level inverter systems.The specific contents include the following four aspects:Firstly,the finite state machine and rouigh set theory-based strategy is proposed for multiple transistor open-circuit fault diagnosis in T-type three-level inverters.Considering the complexity of fault behavior and the difficulty of fault feature identification for multilevel inverters,a finite state machine model of T-type three-level inverter under fault operating mode is proposed to track the state transitions when open-circuit faults occur in single or multiple transistors,and obtain the diagnostics reflecting the undersired state transitons.To simplify the diagnosis strategy and decrease redundant diagnosis information,the rough set theory is employed to optimize and provide a minimum set of sginals necessary to distinguish state transitions.By analyzing the logic relationships voltage state variables in the signal set,a simple diagnosis circuit is designed for fast online fault identification.Also,the circuit can eliminate the effect on diagnosis results due to the signal delay and switching noises.Following simulation and experimental results,the proposed method has the merit of fast diagnosis speed,simple implementation,and strong robustness.Secondly,a graph theory-based modeling method of T-type three-level inverter is proposed to evaluate the influence of short-circuit and open-circuit faults on the operating mode,and overcome the traditional circuit diagnosis model only for specific open-circuit fault conditions,enhancing its versatility.By building the node-path state tracking model of T-type inverters,a voltage vector deviation characteristics-based diagnosis method in accordance with the derivation of path states and node potentials under various fault scenarios.To distinguish the faulty transistors sharing similar features,a path state reconfiguration approach is given to achieve fast fault location.Furthermore,a flexible threshold adaptive strategy is adopted to improve the reliability of diagnosis results.Comparing to existing schemes,the proposed diagnosis method can locate faulty transistor rapidly and accurately without extra hardwares and sampling frequency.Experimental results verify the robustness and effectiveness of the proposed method.Then,a forward-feed convolution network model-based real-time diagnosis method for transistor open-circuit faults of T-type three-level inverters is proposed to eliminate the dependence of diagnosis system design on circuit model and parameters.In this model,a simple auxiliary neural network is used to learn the circuit operating characteristics,and transfer the trained filter paramters for the multilayer convolution feature classifier to enhance its discrimination.To further improve the diagnosis precision,an image construction strategy based on multi-sequence dimension expansion is also proposed to expand the receptive field of local information by transforming the time-series diagnosis signals into two-dimension data feature maps for the feature classifier.In addition,the proposed model is more suitable for real-time diagnosis due to a global pooling operation applied in the classifier to reduce the test time.For the proposed diagnosis method,there is no need to build the precise circuit model or deduce its operating mechanism in detail.The abnormal status of T-type inverters can be identified only by learning the time-domain information involved in three-phase currents,and thus,the proposed diagnosis strategy has the potential to be used in other applications.Through the experiment results,the effectiveness of this approach is proved.Lastly,a sharing redundant unit topology is proposed to maintain the system operation under multiple leg open-circuit fault conditions.The proposed redundant unit is connected by clamped diodes to each middle transistor of the base inverter,and the H-type topology consisting of two paralled legs is formed in each phase.The redundant operation is reachable through the standy legs providing extra current paths.During fault-tolerant operation,the healthy devices in faulty phase are also utilized to reduce the amount of extra hardwares,and the rated output is still possible.To further improve the fault tolerance performance,a recovering three-level tolerant control strategy is proposed on the basis of the redundant topology.By applying the control strategy,the switching sequence can be rearragned by establishing virtual vectors to ensure the predetermined line-to-line voltages and recover the three-level modulation.By adjusting virtual vectors dynamically to suppress the fluctuation and bias of neutral point potentials,the voltage of dc-linck capacitors can be balanced in the scenario of multiple transistor faults.The proposed method combines the merits of existing software-and hardware-tolerant strategis,which effectively enchance the system reliability and improve the output quality.Experimental tests verify the phase current THD being lower due to the shift to partial three-level operation.It shows better promise for reliability-critical and uninterrupted power supply. |
PDF Full Text Request