Over the past years, the advances in artificial intelligence, such as better optimization tools and the enabling of complex Deep Neural Networks by advances in computational resources, has open vast possibilities for Chemical Engineering applications. In this work, four applications of Neural Networks in industrial Chemical Engineering systems that we have worked on are discussed, where the main challenges and advances of artificial intelligence over the past years are reflected: Firstly, an early effort for the development of an Artificial Neural Network (ANN) for the prediction of overall gas holdups in bubble columns is shown. A vast experimental database, which consisted of up to 3500 data points obtained from literature data, was used to develop an Artificial Neural Network (ANN). Such ANN consisted of one single hidden layer and allowed to estimate with reasonable accuracy the overall gas holdup in bubble columns operated at vastly different operation conditions. This ANN also allowed the estimation of the key dimensionless groups that were more relevant in determining the overall gas holdup in bubble columns. However, for design and scale-up tasks of bubble columns, it is needed to predict the time averaged radial gas holdup profiles. In this sense, a more complex neural network was developed for prediction of the time averaged radial gas holdup profiles. Taking experimental data obtained from our in-house developed advanced measurement techniques, such as optical fiber probes, and gamma-ray computed tomography, a data base with 1300 data points was obtained. A Deep Neural Network (DNN) with 2 hidden dense layers was developed and allowed the reliable prediction of the radial gas holdup profiles in bubble columns operated at vastly different operation conditions. Such DNN requires limited computational resources. Furthermore, using the same data base, A second order polynomial model, was developed and simplified, not only to allow for the prediction of the time averaged radial gas holdup profiles within our experimental range but also to provide an insight into the complicated inter-relations between the physical properties and operating and design parameters and thus, the quantification of the effects of these interactions. DNN, in comparison with ANN, have allowed to incorporate more complex architectures, which allow a better prediction of more complex and highly non-linear phenomena. In this context, a DNN was also developed for predicting the local dynamic liquid holdup in a Trickle Bed Reactor (TBR). Predicting local hydrodynamic parameters in TBRs represents major challenge due to the three-phase interactions, and the lack of available experimental measurements of local profiles. A databank was developed by conducting new experimental measurements on a TBR operated at different flow conditions, using our in-house developed gamma-ray computed tomography. This new databank was used to develop the DNN, which incorporated 3 hidden dense layers. The DNN was able to predict the local dynamic liquid holdup in the TBR and outperformed in predictive quality available models reported in literature. A major limitation for the Neural Networks is obtaining a databank that has enough data points to obtain reliable predictions. For example, a complex application is the prediction of the time averaged radial holdup profiles in gas-solid fluidized beds. The available experimental data for local gas holdup in fluidized beds is a limited data environment. Hence, application of a neural network is not possible. In order to apply the deep neural network for fluidized beds, a hybrid approach with validated computational fluid dynamics (CFD) simulations for the development of the data base was developed. In the proposed hybrid approach, an experimental data base, which consisted of around 500 datapoints, was complemented with additional data points obtained from predicted profiles from a validated CFD model at various conditions. This pairing of experimental and predicted data points allowed to increase the data base to 5000 data points. The developed Deep Neural Network consisted of 5 dense hidden layers, and allowed reliable predictions of the experimental profiles, and the exploration of different operation conditions. As depicted on these examples, the application of Machine Learning through the development of Artificial Neural Networks (ANN) for Chemical Engineering systems has emerged as a promising alternative for tackling ever more complex problems. This increasing interest in ANN has been also wide explored in literature and is continuing to be developed. Among the vast applications, ANN are being widely applied for process control, modelling, optimization, and fault-detection tasks, in areas such as energy, chemicals, pharmaceuticals, food and drink, and biotechnology and water. Recently, ANN have demonstrated a high efficiency in simulating continuous stirred tank reactors, design of adaptive neural impedance controllers, modelling and optimization of zinc oxide nanorods synthesis on activated carbon, modelling and optimization of batch fermentation, and optimization of polymerization processes, among others. These general applications in chemical engineering will be outlined.