This research focuses on trustworthy federated learning in complex distributed environments involving non-independent and identically distributed data, heterogeneous client capabilities, malicious model updates, data poisoning, and opaque aggregation decisions. Explainable artificial intelligence techniques are employed to analyze feature contributions, decision mechanisms, and the quality of local model updates. Based on these analyses, interpretable client trust assessment, adaptive weighted aggregation, and anomalous update detection methods are developed. Major topics include robust federated aggregation, client contribution evaluation, model update quality assessment, confidence calibration, federated unlearning, privacy preservation, and model behavior auditing.