Innovative approaches to sludge management: The role of AI, machine learning, and emerging technologies

Sludge can be heavy and unwiedly before it is dewatered and dried, leading to expensive hauling costs or process inefficiencies. Artificial intelligence and newer digital technologies may hold the key to improving this for wastewater systems across the U.S.

In sludge dewatering, free water and some interstitial (floc) water is removed from the sludge resulting in a sludge with dry solids content of between 15-45%. The dewatered sludge has less weight and volume than the original sludge, making it easier to dispose of. This article discusses the use artificial intelligence (AI) for sludge management including for sludge dewatering. 

Sludge conditioning prior to sludge dewatering is important

Before the sludge is dewatered, it is conditioned to break down the chemical bonds so that the water can be released in the dewatering process. In the absence of sludge conditioning, sludge dewatering is not very effective. Chemicals including coagulating and flocculating agents are typically used to aggregate the solids in the sludge to allow the water to be released. 

Factors that affect sludge dewatering

It is important to understand the factors that affect sludge dewatering because dewatering is an integral process for sludge treatment and management. Some of the factors that affect sludge dewatering include the concentration of extracellular polymeric substances, particle sizes, and electric charges of the sludge.

Concentration of extra cellular polymeric substances (EPS)

EPS is a natural polymer secreted by microorganisms, and it generally accounts for 50% to 90% of the total organic mass in activated sludge. EPS adheres to the surface of the sludge particles where bound water is also found. The effectiveness of sludge flocculation, sedimentation, and dewatering are determined by the quantity of EPS in the sludge.

Particle sizes, viscosity and adsorption

The particle size affects viscosity and the adsorption of the water molecules which in turn determine if the particle is capable of water loss. Sludge has a small particle size and thus has a large viscosity and specific surface area. These characteristics make the sludge particle able to strongly adsorb water. 

Electrical charges of the sludge surface

Due to its chemical composition, sludge also has a large number of negative charges on its surface. Because water is a polar molecule, meaning it has a partial negative charge on one side because of the highly electronegative oxygen atom and partial positive charges on the hydrogen atoms, the positive charges on the water molecule are attracted to the negative charges on the surface of the sludge. The greater the number of negative charges on the sludge, the more the number of water molecules bind to the surface and therefore, the more difficult it is to remove water from the sludge. 

Resistance of the sludge to filtration

The higher the sludge resistance to filtration, the more moisture the sludge has and thus, the more the need to find dewatering methods to dewater the sludge. The resistance of the sludge to filtration is affected by the sludge particle size and the surface charge of the sludge, as discussed above. 

A study found that the specific resistance of the sludge to filtration depends on factors including the reagents used for water clarification and the duration of sludge accumulation in the settling tank. 

What technologies exist for dewatering sludge today?

Basic technologies

The basic technologies are filter press, centrifuge, rotatory press, and sludge bed drying. These technologies are also modified to produce other technologies for sludge dewatering. Examples include the hydrocyclones and rotating drum screens. 

Physical technologies

There are also physical technologies. These are the use of ultrasonic techniques, freeze-thaw, and activated carbon. Activated carbon is a porous material with good adsorbance ability. When used on sludge, it adsorbs the water from the sludge to enhance the sludge dewaterability.

Chemical technologies

Chemical technologies include advanced oxidation technologies and the application of hydrogen peroxide. The hydrogen peroxide treatment can be supplemented with other chemicals such as lime and red mud to improve the dehydration of the sludge for better sludge dewatering.

Emerging technologies

Emerging technologies are electrochemical methods that involve the application of electric fields to remove water and bio-dewatering methods that involve the use of microorganisms to disintegrate the organic material in the sludge. This allows water to be released relatively easily from the sludge than without the disintegration of the organic material.

What considerations should a utility manager make before investing in new equipment or processes?

It is essential for utility managers to identify the needs and the challenges for sludge management before investing in new equipment or processes. Some of the key factors utility managers should consider before investing in new equipment or processes for sludge dewatering include:

What training is required for wastewater plant operators?

Plant operators are responsible for ensuring the proper treatment processes including for sludge dewatering, conducting regular maintenance, and troubleshooting issues. As such, plant operators require training to use new equipment or processes before these can be implemented for sludge dewatering.

How can the utility best address its sludge volume?

Due to the increasing urban population, more wastewater is generated and thus more sludge is generated. Utility managers need to identify the volume of sludge at their wastewater treatment plants and which of the new equipment and technologies are best suited to cost-effectively dewater the sludge.

What reuse option exist and how can the sludge be reused?

The chemical composition of the dewatered sludge and the post-treatment for eliminating pathogens are important factors to determine the best reuse practice of dewatered sludge. One of the uses of dewatered sludge is as fertilizer.

What are some common or universal benchmarks that utilities try to achieve with their sludge dewatering processes?

The common or universal benchmark that all wastewater treatment plants attempt to achieve with their sludge dewatering processes is to produce a sludge that has a dry solids content of between 15-45% to make it easier to dispose.

How have technologies evolved for the modern wastewater treatment plant?

Technologies for the modern wastewater treatment plant are evolving towards the use of artificial intelligence (AI). The main objectives in using AI in wastewater treatment plants including for sludge dewatering are threefold.

1. Identify operational parameters that influence the wastewater treatment processes
2. Use these operational parameters in AI to optimize the wastewater treatment processes
3. Decrease the traditional laborious process for wastewater management to handle increased in sludge quantity in the future

From data to artificial intelligence (AI)

There are essentially four steps from data to intelligence and to automated processes: Data, information, knowledge, and AI.

Data: Data is collected from qualitative and quantitative measurements or recordings of a process from studies done in laboratories or fields. The data is used and analyzed in process engineering, automation engineering, and in data science to produce information.

Information: The information is stored and accessed by wastewater management facility operators, engineers, management and administration. It is also used in engineering models.

Knowledge: Knowledge is the comprehension of the information, including those informational points from engineering models. Knowledge provides insight into mechanisms within a certain context. It is used for intelligence and to develop digitalization and automated processes.

Intelligence: The knowledge is fed into intelligence programs and the results are used to optimize wastewater treatment processes including in automated processes for sludge dewatering.

A closer look at AI and machine learning

There are numerous technologies for sludge dewatering. Improving sludge dewatering requires not only improving the core processes involved but also using the generated data in AI. The generated data allows the tracking of critical parameters such as temperature, pressure, and flow rates. When these data networks are coupled with AI-driven analytics, treatment conditions can be optimized.

Artificial neural networks (ANNs) are computer systems designed to mimic the way human brain processes information. They use artificial neurons to analyze data, just like brain neurons. Artificial neural networks (ANNs) are core components of AI, and machine learning (ML) is a subfield of AI.

The following section presents examples of the use of ANNs and ML in sludge conditioning and dewatering. Because sludge conditioning is integral for effective sludge dewatering, several studies discuss the use of ANNs and ML technologies in sludge conditioning.

Examples of the use of ANNs for sludge dewatering

ANNs have been applied in the mechanical dewatering of sewage sludge. This treatment involved sludge physical conditioning using cement and was followed by ultrasonic disintegration. The conditioned sludge was then dewatered by centrifugation. The parameters involved were:

• For sludge physical conditioning, the dose of the cement;
• For ultrasonic disintegration, the maximum amplitude, applied amplitude for sonication, and sonication time; and
• For centrifugation, the relative centrifugal force.

ANNs were applied to select and optimize these parameters. These parameters were used as inputs in ANNs and the output variables were sludge hydration and separation factors. Two non-linear networks were selected and used to forecast the effectiveness of municipal sludge dewatering: a multilayer perceptron and a radial neural network.

The prediction error did not exceed 1% of the real value and ANNs were useful in optimizing sludge dewatering efficiency using the above parameters. However the disadvantage of ANNs is that large datasets, in this case more than 600 research cases, are needed to train the ANN.

In another study, a nonlinear, adaptive, and self-organizing ANN model was applied to predict the dewaterability of waste-activated sludge using multiple physiochemical properties and conditioning operation parameters as inputs. These properties were broken down into three categories.

1. Physical properties: Particle size, zeta potential, and interfacial energy
2. Chemical properties: Alkalinity, extrapolymeric substances, and cell integrity
3. Operating parameters: Type and dosage of conditioning agent, amount of feedstock, and batch run time

These properties were used as inputs to establish the ANN model, and the output was used to optimize the final water content of the dewatered sludge. This approach did not require laborious filtration and simulation tests because the ANN model predicted the dewaterability of the sludge based on the multiple physiochemical properties that were thought to have strong correlations with the sludge dewatering performance.

The results provided a reliable index system for selecting technology and optimizing the process for sludge conditioning that involved sludge conditioning, pressure filtration, and vacuum drying.

Machine learning (ML) to manage the quantity of sludge generated and in sludge conditioning

The quantity of sludge generated and sludge conditioning are two factors that affect sludge dewatering. Regarding the quantity of sludge, the more sludge there is, the more the number of negative charges on the sludge, making it harder for water loss from the sludge. The importance of sludge conditioning is discussed in a preceding section in this article.

An ML approach was developed to predict the amount of sludge generated from a wastewater treatment plant in Ulaanbaatar central wastewater treatment plant in Mongolia besides also being used to predict the daily disposal of wastewater. The ML approach was used according to these steps:

1. Data preparation process: Data on the amount of sludge generated and the amount of wastewater disposed were combined with climate data.
2. Response variables and independent variables were identified.
3. Model fitting experiments were conducted to determine the best fitting model.
4. The model was applied for optimal planning including for sludge management.

Another study developed a ML model, Explainable Boosting Machine (EBM), to classify mixer speed label (fast mixer speed, slow mixer speed, appropriate/suitable mixer speed) based on floc images. In sludge conditioning, mixing the sludge with the chemical polymer produces floc. Appropriate dosing of the chemical polymer and the mixing speed are important factors for effectively mixing the sludge with the polymer to achieve conditioning of the sludge and to eventually improve the efficiency of the dewatering process.

The floc images were used to extract floc gaps and texture. Floc gaps refer to the gaps between flocs, specifically gap radius and number of gaps. Floc texture refers to the overall shape of the floc. EBM was used because it allows interpretation of the input data. The workflow was:

1. Input: Original image on floc that includes gap and texture
2. Feature extraction:
   • Gap information: Gap acquisition, calculation of gap radius, number of gaps, and histogram
   • Texture information: Extraction of gray level differences (it concerns pixels in images due to the texture of the floc), and histogram
3. Learning: EBM
4. Classification: Overmixed, suitable, or undermixed

The study concluded that both features, gaps and texture of the floc, are important input to classify mixer speed labels (overmixed, undermixed, suitable) and if any of these two features are missing in the input, the accuracy of the model is poor for mixer speed labels.

ML for ultrasonic pretreatment of sludge

Using principal component analysis, a study showed that ultrasonic density and ultrasonic time were positively correlated with soluble chemical oxygen demand (SCOD), total nitrogen, and total phosphorus in the sludge. Among five machine learning models, the best model for SCOD prediction was XG-boost while RF was the best model for predicting total nitrogen and total phosphorus.

The study also showed that through SHapley Additive explanation (SHAP), the important variable for SCOD was ultrasonic time while for total nitrogen and total phosphorus, the important variable was sludge concentration. The flowchart consisted of four steps:

1. Data collection: This involved literature survey and screening, data extraction, and excluding unreasonable data.
2. Input and output: Input features were operating conditions (ultrasonic frequency, power, time, and treatment volume) and sludge condition (sludge source, sludge concentration, and pH).
3. Data processing and ML model: The dataset division was 20% test data and 80% train data. This was followed by data conversion and cultivation for use in ML. The ML algorithms used were RR, MLPNN, RF, SVR, and XG-boost.
4. Results: The best ML algorithms were selected and provided in a software package for study applications.

Advanced oxidation processes disrupt the matrix of extracellular polymeric substances, thus allowing better sludge dewaterability. However, the optimization of advanced oxidation processes is challenging because numerous parameters must be identified and tested.

To make this optimization easier, a study integrated advanced oxidative processes with ML to establish a predictive framework for optimizing parameters. The study concluded that a Bayesian-optimized XGBoost model was relatively better than other algorithms to predict optimal configuration for advanced oxidation process while an AdaBoost-based model provided mechanistic insights. The workflow was:

1. Data curation: articles were used for data processing, then encoding, followed by 5-fold cross validation.
2. Model selection: Models included XGBoost and AdaBoost.
3. Model analysis: SHAP (SHapely Additive ExPlanations) and expert-driven analysis were used.

Machine learning (ML) in sludge dewatering

A scholarly article used ML to optimize the pyrolysis process on sludge. The study showed that the use of ML in pyrolysis for sludge dewatering not only optimized sludge dewatering but also maximized energy recovery and minimized carbon emissions.

ML has also been used to optimize the filter press process in sludge dewatering. A study showed that an ML model of support vector regression had the best accuracy in industrial simulation with simulated mean relative error for moisture of 1.57% and 3.81% for processing capacity. This model not only saves energy consumption and reduces operation costs but it also has the potential to improve the efficiency of the filter press process for sludge dewatering. The optimization process was as follows:

1. Feeding the parameters into the model: The condition parameter was the feed concentration. The control parameters were the feed time, squeeze time, and air drying time.
2. ML model: The ML models used were radial basis function (RBF)-orthogonal least square, RBF-generalized regression neural network, and support vector regression.
3. Simulation and prediction accuracy: the simulation and prediction accuracy of the ML models were compared.
4. Obtaining the optimal value and optimal control parameters: The simulation results were sequentially compared according to different optimization principles to find the optimal ML model and the results were used for the sludge filter press method.

What can utilities do to improve or optimize their sludge dewatering processes?

To improve or optimize their sludge dewatering process, utilities need to consider the following issues:

The factors affecting sludge dewatering

They need to consider the chemical composition of the sludge including the concentration of the extracellular polymeric substances because these substances affect sludge flocculation and dewaterability. As explained in this article, the amount of sludge and the electric charges on the sludge are also important factors to consider because they affect sludge dewaterability.

The technologies to use

While basic technologies, namely filter press, centrifuges, rotary presses, and sludge bed drying are commonly used, there are emerging technologies such as electrochemical methods and bio-dewatering. The selection of a technology depends on several factors including the amount of sludge in the wastewater received per day. Financial resources required to train wastewater operators to use the technology and to implement the technology also need to be considered.

Begin working with AI

Selecting a suitable technology could be laborious, and as this article has shown, there is a surge in the use of AI for optimizing the processes for wastewater treatment, including for sludge dewatering. There are ANNs models and ML studies that have shown effectiveness in optimizing the sludge conditioning and dewatering.

Using and implementing AI involves considering the available financial resources. The use of AI is also beneficial because it prompts wastewater facilities to identity the parameters that are important to use in AI for optimizing sludge conditioning and dewatering, keeping in mind that as the population grows, the quantity of sludge received at the wastewater plant will also grow. The current trend is using AI technologies for sludge conditioning and dewatering.