Stateflow models are widely used in industry to implement the high-level control logic of Cyber-Physical Systems (CPSs) developed in Simulink. While extensive research exists on testing Simulink and Stateflow models, repairing detected faults remains largely manual, leading to high debugging costs. Automated Program Repair (APR) techniques can mitigate this cost, but existing approaches face major challenges when applied to CPSs, including long simulation-based test execution times and inadequate repair objectives that fail to capture CPS-specific behaviors. Our IST paper presents FlowRepair, a search-based APR approach specifically designed to repair faulty Stateflow models. FlowRepair introduces a novel repair algorithm that combines global and local search to improve scalability under expensive simulation-based testing. To guide the search effectively, it defines three repair objectives tailored to CPSs: minimizing the duration of failure, reducing failure severity, and delaying the onset of failures. In addition, FlowRepair provides a set of mutation operators specifically designed for Stateflow states and transitions. We evaluate FlowRepair on a dataset of 19 faulty Stateflow models with real bugs, spanning multiple CPS domains. The results show that FlowRepair successfully generates plausible patches for 18 models and valid patches for 16 models, outperforming a strong baseline APR approach inspired by existing CPS repair techniques. These results demonstrate that FlowRepair is both effective and scalable for repairing CPS controllers modeled in Simulink-Stateflow.