Image‐based analysis of fresh concrete flow ‐ determining the correlation between flow behavior and rheological properties

Besides compressive strength, the workability of fresh concrete is one of the most important characteristics of concrete. Only by adapting the rheology of the fresh concrete to the geometry of the element to be casted, the formation of air voids and damages can be avoided. This holds even more so true for 3D printable concrete. Currently, empirical test methods such as the slump test are used to determine rheological parameters. Rheometer tests, on the other hand, allow a much deeper insight into the rheological properties of concrete but are more challenging. Further, all currently available test methods can only be applied on a batch basis.


Introduction
The quality of concrete structures is highly influenced by the rheological properties of the fresh concrete, regardless of whether it is cast using a classical formwork or 3D printed [1][2][3].During casting, unsuitable rheological properties may lead to serious problems like segregation, flow blockage or substantial voids in the structure to be cast [1].The rheology of fresh concrete can be simplified by the Bingham parameters yield stress and plastic viscosity μ, as well as thixotropy (e.g., the change in dynamic viscosity η as a function of time and shear history), which can be determined by rheometer tests [1,4].More specifically, by applying the Bingham model, values for the yield stress and the plastic viscosity µ can be determined with where describes the shear rate in [1/s] and denotes the shear stress in [Pa] [3].
Rheometer tests, however, are laborious, expensive in equipment and difficult to apply on the construction site.Further, they are exclusively batch-based, not allowing for a holistic quality control of the entire delivery.As a consequence, the quality control at the construction site today is carried out using empirical test methods, regulated by norms, from which the workability or the flowability is derived.Therefore, an increasing interest has emerged in determining the rheology at the construction site by, for instance, correlating the energy consumption for rotating the mixing drum of a mixing truck to values obtained by rheometer tests or using a concrete mixing truck itself as a rheometer [5,6].So far, none of these methods has led to an expedient solution.Consequently, there is still a huge demand and high potential for approaches enabling an automatic and real-time on-site concrete characterization and quality control.This is especially true for fresh concrete with increased requirements, e.g.3D concrete printing applications, where a very precise and especially continuous concrete quality control is highly desirable [7].Beyond that, cement reduced eco-concretes are becoming more common in use, as the concrete industry strives to become more sustainable and climate-friendly.The measures taken to reduce the environmental impact, however, also lead to a reduction in robustness of the concretes, e.g. by reacting more sensitively to fluctuations of the water content and the raw materials [8].Therefore, it is necessary to develop new digital approaches for a realtime and continuous concrete characterization in order to improve quality control.
In concrete science, horizontal channels and flow boxes have become a common tool for gaining information about rheological properties [9,10].However, the available studies are exclusively focused on non-stationary conditions so far, i.e. they consider the gradual filling of a container going along with a horizontal, gravity driven levelling of the material.Based on these studies, the yield stress is considered as one of the controlling parameters during form filling [1,10].The plastic viscosity on the other hand is considered as the dominating parameter for the flow velocity and is e.g.frequently measured experimentally using a V-type funnel or a cone [11,12].
The application of optical methods as an additional, quantitative characterization method for fresh concrete shows a high potential.In [13] a camera based characterization method for fresh concrete in the flow table test was developed.In [14] the flow behavior of ultrasound gel was studied in experimental investigations, predicting rheological properties of the material out of the flow behavior using a CNN based evaluation method of recorded stereo image frames.Beyond that, studies were carried out evaluating images of fresh concrete from the mixing process by extracting the shape of the concrete inside the mixer, for determining the slump flow and V-funnel flow time [15].It would thus be highly interesting to continuously monitor the rheological properties of concrete, e.g. during the unloading of a truck mixer utilizing optical measurement methods.
Preliminary investigations of analyzing the flow behavior of fresh concrete during the unloading process of a mixing truck have shown a correlation between rheological properties and flow behavior [16].To ensure constant experimental conditions, an experimental test setup is presented in this paper to conduct a series of experiments under laboratory conditions with cement based mortar.Mortar is much more homogeneous compared to concrete and is used for benchmarking purposes.The setup consists of an inclined open channel ensuring stationary flow, equipped with a video camera system to investigate and analyze the flow behavior using computer vision methods.
In summary, we make the following contributions in this paper: 1.The development of a unique experimental setup consisting of an inclined channel for analyzing the flow behavior of fresh mortar using optical measurement techniques under application of computer vision methods.2. The demonstration of a novel approach for a non-contact measurement method for the determination of the plastic viscosity and yield stress 3. Establishing the basis of a novel approach to continuously monitor the rheological properties of concrete during the unloading process of a mixing truck 2 Materials and Methods

Materials
The laboratory investigations described in this paper were performed using a mortar containing an Ordinary Portland Cement (OPC) CEM I 42,5 R (775 kg/m³; composition and properties see [17]), natural river sand with 2 mm maximum grain size (1161 kg/m³; Weser river sand), with a water to cement ratio w/c of 0.38.Additionally, a retarder (0.8 % by weight of cement (bwoc)) (MasterSet R 436, Master Builders Solutions Deutschland GmbH) and a polycarboxylate-based superplasticizer (MasterGlenium SKY 594, Master Builders Solutions Deutschland GmbH) were added.One mixture was used for all experiments shown in this paper by only varying the superplasticizer content.In total, five different mortars were prepared as summarized in Table 1.0.80 % 1.00 % 1.25 % 1.50 % 1.80 % All mixes were prepared using a compulsory mixer (ZM 100, UEZ Mischtechnik GmbH), with a batch volume of 70 dm³ per mix.The mix was prepared by first homogenizing the dry constituents for 30 s, then adding water and superplasticizer within the next 30 s and then mixing it for another 390 s.

Open channel experimental setup
For the test series, an open channel system was developed as shown in Figure 1.It consists of a rectangular open channel with a length of 120 cm, a width of 10 cm and a height of 12 cm.The inclination angle of the channel was set to 11.5°.To provide a constant influx, the channel was connected to a dosage container with a volume of 90 dm³ and the flow rate was controlled via a manually operated sluice gate that is placed at the channel entrance.Both the dosage container as well as the channel were constructed from film-faced plywood.Furthermore, the open channel was equipped with an obstacle that was installed 75 cm behind the sluice gate in the direction of flow.In the presented investigations, a 3D-printed square shaped obstacle with an edge length of 3.0 cm made from polylactic acid (PLA) was used, thus reducing the width of the channel in the obstacle obstructed cross section to two times 3.5 cm.During the channel flow, the obstacle was engulfed by the mortar while being monitored with videocameras.
Two cameras (GoPro Hero9) with two different field of views (FOV) were installed above the channel, each observing an area of the channel over a length of 40 cm.The first camera was placed 15 cm in front of the obstacle observing the undisturbed flow of the mortar, while the second camera was placed 12 cm behind the obstacle, observing the mortar's engulfing behavior and the confluence of the mortar behind the obstacle.

Test procedure
For the experiments, the following test procedure was applied: In a first step, 70 dm³ of cement mortar were mixed corresponding to the composition of sample M_0.80 (cf.Table 1) and filled into the dosage container.The image acquisition was started and the sluice gate was lifted 1 cm, causing the material to flow down the channel.At the end of the channel the material was recollected and reference samples of the mortar were taken and tested for their rheological properties using a rheometer (cf.Sec.2.4).Subsequently, an amount of superplasticizer as specified in Table 1 was added to the mortar (i.e. to the mix M_0.80) in order to adjust the composition to correspond to blend M_1.00; homogenization was achieved using a twin-shaft mixer.The resulting new mix was subsequently filled back into the dosage container to start the next measurement.Between each measurement the channel was cleaned.The described test procedure was repeated for all following blends M_1.25, M_1.50 and M_1.80, using the initial test volume and varying the superplasticizer content.The experimental procedure was conducted two times, resulting in two test series.

Optical Measurements
The optical measurements were carried out utilizing two video cameras, recording with a frame rate of 120 frames per second (FPS) and an image size of 1920 x 1080 pixels (px).Both cameras were placed at 40 cm above the channel, resulting in a spatial resolution with a ground sampling distance (GSD) of 0.27 mm per pixel at the mortar surface.
For data evaluation a region of interest (ROI 1) with an image size of 531 x 431 px was extracted from FOV 1 to investigate the undisturbed flow in front of the obstacle.
The approach for dense optical flow computation proposed by Farneback [19] was used in order to estimate a displacement vector , for each image coordinate , , i.e., for each pixel, describing the pixel's movement in the image plane between two consecutive image frames.Under the constraint of affine motion in a neighborhood, the actual velocity for each pixel can thus be calculated as follows: As a result, the dense velocity flow field of the mortar's surface flow velocity could be computed for every image frame as exemplarily shown in Figure 2. The dense velocity fields were computed as described for each image frame as well as for each mortar sample and served as foundation for the experimental evaluation.In order to analyze the engulfing and confluence behavior behind the obstacle, a second region of interest (ROI 2) was extracted with identical size as ROI 1 from FOV 2 of the second camera, analogous to the procedure conducted for the extraction of ROI 1.As exemplarily shown in Figure 9 (a) a triangle-shaped flow-gap directly behind the obstacle could be observed.Moreover a surface gap occurred, as the two separated mortar streams did not completely merge (cf. Figure 9 (a)).For data evaluation the images were in a first step converted to binary images to detect the edges of the surface gap (cf. Figure 3).The average distance b between the two edges was then calculated, which represents the width of the surface gap (cf. Figure 3).This value was further used as the parameter for the evaluation.

Rheological Reference Measurements
As reference values, rotational-controlled rheological measurements were carried out with a rheometer for each sample.The rheometer (Viskomat NT, Co. Schleibinger Testing Systems) is a couette type, whereby the container and sample rotate while the torque sensor is static.A paddle specifically for mortar was used as the measurement geometry.The shear profile used is shown in Figure 4.It consists of a steady state phase of 30 s with constant rotational velocity followed by a stepwise decrease of rotational velocity.The data evaluation was carried out using the Bingham model, determined by a linear regression of the data points between 10 rpm and 60 rpm.The absolute values for yield stress and plastic viscosity were determined using the affine-translation approach for arbitrary geometries, presented in [18].The rheometer tests were conducted simultaneously to the open channel experiments by taking a small sample out of the channel.In addition the Haegermann slump flow on a dry glass plate without shocking was carried out right after starting the rheometer test.The results are presented in the following chapter.
Figure 4 Rotational controlled rheometer profile with stepwise decrease in rotational velocity.

Results and discussion
In this section, the experimental results of the rheological and optical rheological measurements are presented and discussed.All experimental investigations were conducted two times using the same mix design and test procedure as described in chapter 2. The mean values of the two test series were computed, with error bars showing the lower and upper bounds.In particular, we investigate the following questions: -What are the rheological properties of the investigated mortars determined by rheometer batch testing (see Sec.

Rheological properties
The addition of superplasticizer resulted in an increasing flowability of the mortars, reflected in a reduced yield stress and plastic viscosity, while the slump flow increased.The results of the slump flow and of the (reference) rheometer measurements are shown in Figure 5.

Engulfing and confluence behavior behind obstacle
By incorporating a prism-shaped flow obstacle in the flow channel, a triangle-shaped flow-gap directly behind the flow obstacle could be observed.The flow pattern behind this obstacle differed in form and length as well as resulting surface gap with a different width (distance b between the two separated mortar streams), depending on the mortars rheology (definitions see Fig. 3).In the course of the experiments it could be observed, that the width of the resulting surface gap increased with a decreasing flow velocity.For the evaluation of this phenomenon, the flow pattern was quantified optically at the point in time, when each mortar had a flow velocity of 0.04 m/s.This represents the same flow velocity as mentioned previously, where it was considered as right before stoppage (cf.Sec.3.2).At this very low flow velocity it can be assumed, that the yield stress acts as the controlling parameter describing the final shape and width of the surface gap respectively [1,10]..80 the surface gap completely vanishes when the mortar is arrested, the mortars with decreasing flowability do not completely merge behind the obstacle, leaving a noticeable surface gap.Also, the width of the gap (i.e.distance b, cf. Figure 9) differs at varying rheology and visually appeared to increase in width with decreasing flowability.The average width (cf. Figure 9) of the resulting surface gap for the samples M_0.80, M_1.00 and M_1.25 was then measured.It was then normalized to the edge length of the obstacle used (3.0 cm) and related to the corresponding reference yield stress (cf. Figure 10).Thereby the surface gap increases with higher yield stress.However, here it has to be noted, that the procedure does not seem to be suitable for very fluid mixes.

Conclusion
In this paper, a new computer vision-based approach was presented, to identify typical flow patterns in an inclined open flow channel of different cement mortars by varying their rheological properties.The flow behavior is investigated using a camera system and dense optical flow computations.The results show a direct correlation between the mortar's measured reference rheology and its flow behavior derived from the recorded video data via image analysis.
A stepwise increase of the flowability of the investigated mortars by adding superplasticizer resulted in a higher maximum flow velocity of the mortars in the open channel flow and a decreasing total flow time respectively.Both, the optically derived maximum flow velocity, as well as the measured total flow time, highly correlate to the mortars Bingham plastic viscosity, as it is the dominating parameter on flow velocity.
By investigating the engulfing and confluence behavior behind an obstacle, a clear dependency on the Bingham yield stress of the material could be observed.The experiments demonstrated, that the width of surface gap increases with higher yield stress.For further investigations, evaluating the yield stress out of the engulfing and confluence behavior while the mortars are actually flowing is highly desirable.For this purpose, the shape of the triangle behind the obstacle as well as the distance between the obstacle and the merging point at different flow velocities represent promising parameters.
In the future, a similar setup could be used to investigate the flow behavior of concrete, which in principle can be used for analyzing the unloading process of a mixing truck at the construction site, providing an entirely, digital method of quality control.

Figure 1
Figure 1 Experimental setup: Open flow channel, equipped with a camera system, square shaped obstacle and an inclination angle of 11.5°.

Figure 2
Figure 2 Two-dimensional flow field of the computed surface velocity of undisturbed flow.

Figure 3
Figure 3 Original image with marked edges of the surface gap (left) and corresponding binary image (right), used to detect the edges of the surface gap and to calculate the average width b of the surface gap (distance between dotted line).

2 . 5 )
, as these serve as reference values for the optical measurements.(Sec.3.1) -How does the general flow behavior look like in the open channel, regarding surface flow velocity profile, maximum flow velocity and development of the flow velocity over time.(Sec.3.2) -How does the obstacle influences the mortar's flow behavior, regarding engulfing and confluence behavior behind the obstacle.(Sec.3.3)

Figure 5
Figure 5 Results of investigated mortars for a) slump flow in [mm], b) rheometer yield stress in [Pa], c) rheometer plastic viscosity in [Pa•s] as a mean value from 2 repetitions.

Figure 6
Figure 6 Development of the open channel (OCF) flow velocity [m/s] over flow time [s] for all samples, exemplary for the first test series.

Figure 6
Figure6shows the development of the open channel flow velocity over the time, i.e. the duration of the flow until the dosage container is empty.The initial time at zero seconds is not equal to lifting the sluice gate, rather it represents the point in time right after the flow front of mortar passed the FOV 1, as the flow velocity was computed out of the image frames from camera 1.As can be seen, the hydrostatic pressure inside the dosage container resulted in a high initial flow velocity that decreased in the course of the experiment.Immediately after the opening of the sluice gate (i.e.flow time < 10 s) all investigated mortars show a strongly non-linear behavior, before the flow velocity vs. time curves enter a linear phase until the flow yields at about 0.04 m/s (cf.grey-marked area in figure6) before it stops.

Figure 7
Figure7shows the correlation between total open channel (OCF) flow time and corresponding reference plastic viscosity.Here the point in time at which a marginal flow velocity of 0.04 m/s was reached was defined as the end of flow.In Fig.8then the maximum flow velocity is plotted as a function of externally measured plastic viscosity.As can be seen, both flow parameters show to be in a linear relationship with the plastic viscosity of the material and can thus serve as a good indicator for determining the mortars plastic viscosity from open channel flow investigations.

Figure 7
Figure 7 Mean values of the OCF total flow time [s] until a flow velocity of 0.04 [m/s] was reached and corresponding reference plastic viscosity [Pa s].

Figure 8
Figure 8 Mean values for the OCF maximum flow velocity [m/s] and corresponding reference plastic viscosity [Pa s].

Figure 9
Figure 9 Confluence behavior behind the obstacle representing a triangle-shaped flow gap followed by a permanent surface gap with different widths b for samples a) M_0.80, b) M_1.00, c) M_1.25 and a merged surface for samples d) M_1.50, e) M_1.80 of the first test series.

Figure 10
Figure 10 Normalized width b of the resulting surface gap [-] behind the obstacle at a flow velocity of 0.04 m/s and reference yield stress [Pa].

Figure 9
Figure9exemplarily shows the state behind the obstacle for the samples of the first test series.While for the very fluent mortars M_1.50 and M_1.80 the surface gap completely vanishes when the mortar is arrested, the mortars with decreasing flowability do not completely merge behind the obstacle, leaving a noticeable surface gap.Also, the width of the gap (i.e.distance b, cf.Figure9) differs at varying rheology and visually appeared to increase in width with decreasing flowability.The average width (cf.Figure9) of the resulting surface gap for the samples M_0.80, M_1.00 and M_1.25 was then measured.It was then normalized to the edge length of the obstacle used (3.0 cm) and related to the corresponding reference yield stress (cf.Figure10).Thereby the surface gap increases with higher yield stress.However, here it has to be noted, that the procedure does not seem to be suitable for very fluid mixes.

Table 1
Mortar mixtures, sample name and superplasticizer content (SPC) in percent by weight of cement (bwoc); composition see text