Despite the widespread use of k-means time series clustering in various domains, there exists a gap in the literature regarding its comprehensive evaluation with different time series normalization approaches. This paper seeks to fill this gap by conducting a thorough performance evaluation of k-means time series clustering on real-world open-source time series datasets. The evaluation focuses on two distinct normalization techniques: z-normalization and NP-Free. The former is one of the most commonly used normalization approach for time series. The latter is a real-time time series representation approach, which can serve as a time series normalization approach. The primary objective of this paper is to assess the impact of these two normalization techniques on k-means time series clustering in terms of its clustering quality. The experiments employ the silhouette score, a well-established metric for evaluating the quality of clusters in a dataset. By systematically investigating the performance of k-means time series clustering with these two normalization techniques, this paper addresses the current gap in k-means time series clustering evaluation and contributes valuable insights to the development of time series clustering.