Predictive maintenance (PdM) holds great promise for minimizing unplanned downtime and reducing maintenance costs. However, its adoption remains limited due to the technological complexity of anomaly detection models and the lack of interpretability that undermines user trust. This paper presents an automated, explainable framework for time-series anomaly detection designed to overcome these barriers. Combining a diverse ensemble of models—including one-class SVM, Isolation Forest, autoencoders, and machine learning-based stacks—the framework streamlines model selection, handles dataset variability, and improves detection robustness. A key feature is its emphasis on explainability through an interactive labeling interface and a synthetically generated evaluation dataset, allowing domain experts to explore, validate, and trust model outputs. Experimental results on both real-world photovoltaic energy data and synthetic benchmarks demonstrate that the ensemble-based approach achieves high detection accuracy, even with limited labeled data. By making advanced analytics more interpretable and user-friendly, the proposed framework advances the practical deployment of PdM solutions across industries. The proposed framework is applicable, though not limited, to solar photovoltaic (PV) farms, as demonstrated in the presented case study. Its primary objective is to lower the barrier to entry for implementing machine learning-based predictive maintenance, with a particular emphasis on anomaly detection in time series data.