Reclaimed water for various purposes is a suitable and cost-effective option to compensate for water shortages, preserve existing water resources, and prevent water loss and environmental pollution [1]. Iran is increasingly experiencing scarcity because of climate changes, reduced rainfall, inappropriate management and wasting water. Moreover, Iran, as a developing country, faces rapid industrial development, leading to high water demand in industrial sectors. Thus, it is evident that replacing alternative water sources like recycled wastewater is paramount for industrial usage. Typical industrial water uses include process water, boiler, cooling tower, cleaning and stripping agent. The quality of industrial wastewater depends on the kind of industries, including mining, food and agriculture, tanneries, refineries and pharmaceuticals. It can have toxic contaminants like ammonia, heavy metals, phenols, organic contaminants, solvents and other materials [2,3]. In addition, the quantity of recycled water usage varies from process to process of industries and needs substantially higher quality for boilers and cooling towers, especially in pharmaceutical industries.Given the growing use of treated wastewater and environmental concerns, it is necessary to assess the risk of industrial wastewater treatment plants to reach standard water quality [4]. If the recycled water does not meet the required standard, it can cause detrimental impacts on industrial facilities like sedimentation, fouling, corrosion, scaling and biofilm formation in the pipes and equipment of the industries.In a WWTP, various factors can cause degraded performance, considered a risk of not fulfilling the effluent requirements for reuse purposes. Hence, it is pivotal to analyze distinct failure modes and their impacts on a WWTP through a systematic risk-based approach [5,6]. There are notable risk factors related to an industrial WWTP, which can fall into different categories: equipment failures, human errors, design failures and wet weather conditions. Given that a WWTP has complicated parts and various dynamic performances and faces operational challenges, static models are insufficient for risk analysis. Therefore, a dynamic risk assessment of a WWTP is needed to identify and mitigate system errors and failures to increase system reliability and treated water quality [7].In the current literature, fault tree analysis (FTA), bow-tie (BT), and Bayesian network (BN) are used for risk assessment of water and wastewater treatment plants. The FTA is a deductive approach to determining the potential causes of an adverse event. Beauchamp et al. developed a method for technical and operational hazard identification of the water treatment plant based on FTA [8].For the operational assessment of a reverse osmosis system in a water treatment plant, Bourouni used FTA and reliability block diagram for risk assessment of the reverse osmosis plant [9]. Taheriyoun et al. employed the FTA and Monte Carlo simulation to evaluate the reliability of a WWTP. It was determined that the most significant contribution to the system failure was human errors, climatic and mechanical factors, and sewer system problems, respectively [10]. Piadeh et al. proposed the combined FTA and event tree analysis (ETA) for reliability analysis of advanced treatment unit alternatives. The lowest failure probability was found for the coagulation-flocculation units, while reverse osmosis and ozonation units had the highest failure probabilities [11].Tabesh et al. used two methods, including fuzzy fault tree analysis (FFTA) and Monte Carlo-based FTA, to evaluate the risk of a water treatment plant in Tehran. They found inappropriate reservoir design, power equipment failure and transfer pipe failure as the most effective factors in the plant’s risk [12]. While FTA provides good performance to assess a hazard probability, it cannot describe the consequences. The event tree analysis (ETA) is capable of showing the consequences. Nonetheless, it cannot show how the initiating events’ failure probability has been estimated. The bow-tie (BT) method can overcome the weakness of FTA and ETA by integrating both. It is composed of a fault tree, recognizing the possible basic events causing the critical event, and an event tree, representing the potential consequences of the critical event based on the failure or success [13].Analouei et al. evaluated the risk of an industrial WWTP using the BT. They calculated the WWTP risk by about 41% and found operator errors as the most critical risk factor [14]. Tušer & Oulehlová applied the BT method to identify the causes and consequences of risks in a WWTP based on documentation reviews, safety audits, interviews and check-list methods. The results showed that the low level of safety measures and the age of the WWTP technology could lead to unacceptable and undesirable risks [15].Although FTA, ETA and BT have contributed significantly to the WWTP risk assessment, they cannot show conditional dependency between the variables and the dynamic nature of a system. Moreover, all these tools suffer from uncertainty handling problems [16]. In this regard, the BN is a method that can solve dependence modeling and uncertainty handling problems. It is a probabilistic reasoning network of conditional probabilities [17]. Kabir et al. estimated the risk of an urban water distribution system using BN-based data fusion. The goal of data fusion was to obtain a lower prediction error and higher reliability by using data from multiple distributed sources. The proposed technique was used to identify vulnerable and sensitive water pipes [18].Anbari et al. proposed a risk assessment BN model to prioritize sewer pipes inspection by identifying high-failure risk areas. The results showed that about 62% of sewers had moderate risk while 12% were critical [19]. Zarei et al. mapped the BT method into the BN for the risk assessment of a gas transmission system. The proposed framework was used to evaluate the risk and determine the most critical accident scenarios in the system [20]. Shafiee Neyestanak and Roozbahani investigated the risk assessment of treated wastewater using a novel hybrid BN. The method was applied for risk analysis of Iran’s 27 wastewater treatment plants in four sectors: agriculture, landscape irrigation, groundwater artificial recharge and industry. The results demonstrated the capability of the method to predict the risks, and the risk of using treated wastewater in agriculture (26.9%) was the highest compared to other sectors [21].The conventional BN is static and cannot capture time dependency. Hence, the risk analysis methods mentioned above are not efficient for dynamic risk assessment. In other words, they cannot use time-dependent information to update the probability of events and consequent risk profile [22,23]. Therefore, it is more viable to develop a dynamic Bayesian network (DBN)-based methodology for dynamic risk assessment of WWTPs to overcome the existing methods’ limitations. It will help to understand the time-dependent behaviors of risk factors and ensure the quality of water reclamation.The DBN is an extension of the conventional BN that helps dynamically analyze a system’s risk. It is possible to obtain the trend of the variables’ failure probabilities across the period using the DBN [23,24]. The DBN-based approaches have successfully been used in the risk assessment of different systems and facilities in previous studies. Dawsey et al. developed a methodology based on DBN for real-time water distribution system’s state parameter estimation. They presented an approach for drinking water monitoring using DBN to infer knowledge about the current state of a water distribution system [25].Li et al. used the DBN in the risk analysis of an earth-rock dam breach [26]. Chen et al. conducted a multi-reservoir system’s risk assessment. The authors utilized the DBN to predict the probability of wellhead fatigue failure during the service life [23]. Fam et al. applied DBN for analyzing well-decommissioning failures and long-term monitoring of decommissioned wells [24]. Kammouh et al. proposed an approach based on the BN and DBN to assess the time-dependent resilience of engineering systems [22]. Liu et al. showed how the DBN could be effectively used to model a WWTP. In this regard, a fuzzy partial least squares-based dynamic Bayesian network was presented to improve the prediction of the quality indices in WWTPs when confronting uncertainty, nonlinearity, and time-dependent characteristics [27]. Zhang et al. proposed a model integrating variable importance in in projection with dynamic Bayesian networks (VIP-DBN) to reach better prediction results, optimizing current approaches to handling dynamic characteristics, nonlinearity, and uncertainty simultaneously in papermaking WWTPs. The method was evaluated through two case studies. They showed that the model is an accurate and reliable approach to dealing with the shortcoming of existing soft sensor methods [28].There are other risk assessment techniques that are used to risk analysis of diverse systems. Yari et al. used Data Envelopment Analysis (DEA) to evaluate blasting patterns in mining. In fact, incorrect blasting patterns can result in many safety problems. This study showed the ability of the DEA to blast operations [29]. Yu et al. proposed a comprehensive industrial wastewater treatment plant project evaluation approach based on the improved entropy– Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) method. A case study in china was evaluated using this method to solve the difficulties in accurate quantization and objective evaluation of an industrial sewage treatment project [30].Yari et al. proposed a structure to evaluate the hazards of main decorative stone quarrying by implementing the Preference Ranking Organization Method. They found that economic, management and schedule risks are the most intimidating hazards in this field [31]. Ekmekcioğlu et al. produced district-based vulnerability, hazard and flood risk maps for Istanbul with a hybrid fuzzy analytic hierarchy process (AHP)-TOPSIS model [32]. Yari et al. presented a comprehensive model for ranking mines in the sense of all imposing attributes with an emphasis on safety parameters. In this paper, mines have been ranked using AHP-TOPSIS and fuzzy environment [33]. Hayaty et al. used the fuzzy Delphi analytic hierarchy process and TOPSIS method to risk assessment of Co, Cu, Mo, Zn, Cr, Mn, Ni, Pb, Ti, and Fe in the mining sediment released to the tailings dam. In this research, the metals like copper, iron, and zinc had the highest pollution and critical risk [34]. Lane et al. used a sanitation safety plan (SSP) framework for risk assessment of wastewater treatment systems. This framework identified potential hazards in 29 First Nations wastewater systems in Atlantic Canada. They found that 7% of hazardous events were high-risk while 69% had an unknown level of risk [35].

From the above literature review, it is evident that the current progress in the failure risk assessment of WWTP is noteworthy. However, it lacks a methodology that can handle complex interdependence among contributory risk factors, their multistate modeling, uncertainty, and the dynamic nature of risk and reliability. Therefore, this work aims to overcome these gaps. Since the BN does not accurately capture dynamic systems’ behavior, the DBN can be used to model this temporal process’s nature as a time-dependent probabilistic reasoning approach. Therefore, it can give an accurate estimate of the dynamic risk of a system. The current authors feel it is worth examining DBN’s efficacy in dynamic risk assessment in a WWTP to bridge the existing knowledge gaps and propose a DBN-based methodology for risk assessment and management of industrial wastewater treatment and reclamation plants.

In this work, the risk factors were first identified through a comprehensive evaluation with the help of process flow diagrams (PFDs) and industrial experts, followed by developing the network structure and quantifying the probabilities. The overall study considered an operation for 15 years (2016–2030). The first six years’ data were used to develop the DBN model, and the smoothing inference was used to identify the crucial risk factors. Based on these results, a prevention strategy was developed to effectively reduce the risk factors, which was further applied to the remaining nine years of operation. The results suggest significant improvement in failure risk reduction when the suggested measures are employed.