This paper proposes an LSTM-based Encoder-Decoder (EncDec-AD) model for multi-sensor anomaly detection. The model learns to reconstruct normal time-series behavior and uses reconstruction error to detect anomalies. It is tested on three publicly available quasi-predictable time-series datasets (power demand, space shuttle, ECG) and two real-world engine datasets with both predictable and unpredictable behavior. The model is shown to be robust and effective in detecting anomalies from predictable, unpredictable, periodic, aperiodic, and quasi-periodic time-series, including short and long time-series. The model is trained using only normal sequences, which is particularly useful when anomalous data is not available or is sparse. This approach is especially effective for machines that undergo periodic maintenance, as anomalies may not be present in the sensor data until after maintenance. The EncDec-AD model uses an LSTM encoder to learn a fixed-length vector representation of the input time-series and an LSTM decoder to reconstruct the time-series. The reconstruction error is then used to compute the likelihood of an anomaly at each point. The model is trained to minimize the reconstruction error between the original and reconstructed time-series. The model is evaluated on four real-world datasets: power demand, space shuttle valve, ECG, and engine. The results show that EncDec-AD outperforms existing models in detecting anomalies in both predictable and unpredictable time-series. It is particularly effective for Engine-NP, where the time-series are unpredictable. The model is also able to detect anomalies in time-series with lengths as large as 500, indicating that the LSTM encoder-decoder is learning a robust model of normal behavior. The paper also discusses related work, including time-series prediction models and non-temporal reconstruction models. It concludes that the LSTM-based encoder-decoder approach is a viable method for detecting anomalies in time-series data, even when the time-series is unpredictable.This paper proposes an LSTM-based Encoder-Decoder (EncDec-AD) model for multi-sensor anomaly detection. The model learns to reconstruct normal time-series behavior and uses reconstruction error to detect anomalies. It is tested on three publicly available quasi-predictable time-series datasets (power demand, space shuttle, ECG) and two real-world engine datasets with both predictable and unpredictable behavior. The model is shown to be robust and effective in detecting anomalies from predictable, unpredictable, periodic, aperiodic, and quasi-periodic time-series, including short and long time-series. The model is trained using only normal sequences, which is particularly useful when anomalous data is not available or is sparse. This approach is especially effective for machines that undergo periodic maintenance, as anomalies may not be present in the sensor data until after maintenance. The EncDec-AD model uses an LSTM encoder to learn a fixed-length vector representation of the input time-series and an LSTM decoder to reconstruct the time-series. The reconstruction error is then used to compute the likelihood of an anomaly at each point. The model is trained to minimize the reconstruction error between the original and reconstructed time-series. The model is evaluated on four real-world datasets: power demand, space shuttle valve, ECG, and engine. The results show that EncDec-AD outperforms existing models in detecting anomalies in both predictable and unpredictable time-series. It is particularly effective for Engine-NP, where the time-series are unpredictable. The model is also able to detect anomalies in time-series with lengths as large as 500, indicating that the LSTM encoder-decoder is learning a robust model of normal behavior. The paper also discusses related work, including time-series prediction models and non-temporal reconstruction models. It concludes that the LSTM-based encoder-decoder approach is a viable method for detecting anomalies in time-series data, even when the time-series is unpredictable.